DSBA

Project
List
A comprehensive collection of the
top projects done by our learners
from various domains.
 1.1 Optimization of Equity Portfolio

Portfolio optimization is a formal approach to making investment decisions across a collection of

financial instruments or assets. The portfolio optimization problem lies in minimizing the risk of an
investment for a desired level of expected return. It is estimating asset return and total return moments
from price or return data. The idea behind an “optimal portfolio” comes from the modern portfolio
theory wherein this theory assumes that investors focus their efforts on minimizing risk while also striving
to attain the highest possible return. The business problem here is to provide investment advisory to a
client basis his risk profile and his expectation of return. Advisory is limited to equity asset class only, that
too, for Indian Market. In this project, data from NSE 500, moneycontrol.com and
valuresearchonline.com has been used.
As part of the project, we have done selection of 50 stocks from NSE 500 and have assigned relative
ranking to them based on the financial and fundamental data. Further, we have used non-linear
optimisation techniques and used 10 of these 50 stocks to create a stock portfolio for optimising
returns within a required maximum loss. We have done prediction modelling for stock price which
would be able to predict the stock price for t+3 days and t+10 days fairly accurately. We have used the
techniques of linear, Ridge and Lasso regression and non-linear programming in the project. The
prediction model for stock price has been done using the programming language ‘R’ and the domain of
the project is Finance Risk Analytics.

Keywords: Portfolio Optimization, Non-linear programming, Finance Risk Analytics

1.2 Analytics for a Retail Fashion Brand

The purpose of this capstone is to assess the practical usage of analytical techniques in solving a real-
world business problem. This study has analysed various facets of a retail fashion brand with an objective
to increase its presence and profitability in a highly competitive fashion industry. The brand has a wide
range of progressive seamless wear to maximize flexibility, mobility and performance while enhancing
comfort and confidence. The Management has restrained us from disclosing/ using the name of the
brands and the Company in any kinds of reports/ presentations whatsoever. The vision of the brand is to
symbolize the multiple facets of women’s life which she seamlessly handles in different roles of her life.
Aim of each product of the brand is to truly bring alive the philosophy of “Born Free” with fabric and
design that makes it a fashion statement and even a personal statement. The said brand is a new entrant
in this domain and in a very adolescent stage where it needs a proper and a guided approach to be
amongst the best and leading women’s fashion brand. The fashion industry is very competitive and to
make a presence, a smart approach is required along with the hard work. To compete with the
humungous fashion brands available today, the Company has decided to approach analytical methods to
get the best out of the business. Analytics can help them unearth the insights that would have otherwise
remained hidden. This project will help them make informed decisions that will ultimately shape the
growth of the business and it will help us to understand efficacy of using analytics in real time data.
During the project, three tools were pre-dominantly used viz. Microsoft Excel, Tableau and R. MS Excel is
mainly used for cleaning the data and understanding different variables. Tableau is widely used for
Exploratory Data Analysis and gathering visual insights. R is primarily used for statistical modelling.

Keywords: Retail Fashion Brand, Flexibility, Mobility, Real Time Data, Statistical Modelling

1.3 Sales Analytics on certified pre-owned cars in U.S

Cars are being sold more than ever. Countries adopt the lease culture instead of buying a new car due to
affordability. Therefore, the rise of used cars sales is exponentially increasing. Car sellers sometimes take
advantage of this scenario by listing unrealistic prices owing to the demand. Therefore, arises a need for a
model that can assign a price for a vehicle by evaluating its features taking the prices of other cars into
consideration. In this project, we have use supervised learning method namely Random Forest and
Regression technique to predict the prices of used cars. The
model has been chosen after careful exploratory data analysis to determine the impact of each feature
on price. A Random Forest with 500 Decision Trees were created to train the data. From experimental
results, the training accuracy was found out to be 74 %, The model can predict the price of cars
accurately by choosing the most correlated features.

Keywords: Exponential Increase, Supervised Learning, Random Forest, Regression

1.4 Prediction of policy buyers and their segmentation

An insurance policy lapse is a major concern for most of the insurance companies as it generally occurs
within the first policy year and prevents insurers from recovering initial expenses of the policy
acquisition. Insurance companies have started focusing on retaining the customers, reducing the lapse
rates and hence categorizing them and creating a specific set of customers depending on the various
factors as per the policy outlined by each insurance company. The focus on any cross sell would be only
on these specific customer profiles due a heavy cost of acquisition. A customer’s life cycle in the
insurance business domain depends on the factors like age, gender, marital status, income, insurance,
density of the place etc., Analytics can be used to assess a customer’s total life time value. Here we would
need to achieve the following objectives:

a) Predict if the customer would be willing to buy or not buy the policy.
b) Segment these customers with into proper class like Platinum, Gold, Silver.

By doing the above, we hope that this will increase their product sell and in-turn make their marketing
efforts impactful and effective. In any business, the return on investment is a key metric for assessing a
product profitability.

Keywords: Policy Lapse, Customer Retention, Product Profitability

1.5 Prediction and the rule of law- justice simplified

This project focuses on a class of “Machine Learning” techniques and their potential impact in the domain
of Legal Analytics. This work in no way to be considered as a suggestion that all or most of the tasks
routinely performed by attorneys are automatable. The tools and techniques used to substantiate our
findings include working on R Studio and using Data Mining, Predictive Modelling and Machine Learning
Techniques limited to use of Decision Tree, Logistic Regression, Multinomial Logistic Regression and K
Nearest Neighbours (KNN). We have limited our study to last 20 years’ data and 2 decision variables: Case
Disposition and Decision Type. The findings of this study gives the insight that KNN is one of the judicious
technique for the treatment of missing values, in case of multi-dimensional data. We found out that
Decision Tree algorithm accuracy, precision and recall is better than Logistic Regression and Multinomial
Regression for both the decision variables. Cross Validation technique is also used to verify the
consistency of the model performance. At the end we have been successful to find the core features for a
base model in the Legal Analytics domain, which can be extended further for a deeper understanding of
the legal domain and building a robust and sustainable model.

Keywords: Machine Learning, Legal Analytics, Data Minding, Predictive Modelling, KNN, Cross Validation
 1. 6 Market Penetration – An Analytical Approach

The project is about a body wash product ‘ThinkSkin’ which is a private label for a leading retail
organization. Brand currently enjoys 16% of market share. The organization wants to set a penetration
guidance of 25% in the industry, based on the past purchase patterns and the type of customer Walk-in.
The objective of this project is to develop an analytic framework to generate actionable business
recommendations to enable growth of the category and become market leader in the near future.

Keywords: Retail Organisation, Market Share, Penetration Guidance, Analytic Framework

1.7 Portfolio Risk Assessment and Analysis

In this report we try to understand the relation between Stock Risks ‘β’ in a portfolio, investors Risk
tolerance level ‘λ’ & how investors have invested accordingly. We start the problem by understanding the
risks involved in the stock held by the investor, followed by their risk tolerance assessment and finally by
understanding the relationship between the two metrics. The Markowitz Efficient Frontier inspired us in
understanding that, by simply assessing the stock or market risk is not enough to have better
understanding of the customers behavioural tendency with fluctuations in market. We thereby, illustrate
the underlying relationship in a tabular format to get a good view of the problem.

Keywords: Stock Risks, Risk Tolerance, Markowitz Efficient Frontier

1.8 Marketing Analytics for Bajaj Allianz Life Insurance

Bajaj Allianz Life Insurance Ltd undertakes regular email campaigns to increase brand awareness, sale of
insurance policies and to create unique customer touch points. The core objective of the study is to use
various analytics techniques to deep dive into the responses and aim to optimize the hit rate of these
campaigns. Techniques used here are Exploratory data analysis techniques, Classification algorithms
including Decision Trees and Logistic, Machine learning techniques like Clustering, Neural Network, CART,
SMOTE, Random Forest, Gradient Boosting and Under sampling and oversampling.

Keywords: Marketing analytics, Customer analytics, Random Forest, CART, Neural Networks, Gradient
Boosting and Clustering

1.9 Predicting Fate of Indian Start-Ups

India has the third highest number of start-up incubators and accelerators in the world, after China
and the US. The number of start-ups and incubators grew by 40 percent in 2016-17. Of these, 30
academic incubators were established under the government’s ‘Start-up India, Stand-up India’ initiative.
The idea was simple-simplifying compliance and roll out an action plan for giving a fillip to innovation and
entrepreneurship in the country. In this direction, the government announced measures such as
compliance based on self-certification, a mobile app to provide on-the-go accessibility for registering
start-ups with relevant agencies of the Government, creation of a Start-up India hub, legal support and
fast-tracking patent examination at lower costs, tax exemption for start-ups, providing funding support
through a Fund of Funds, and many other initiatives. Government support, technology enhancement and
development in metro cities has boosted the start-up ecosystem. This has attracted numerous numbers
of investors, both national and globally. Therefore, a large amount of money is poured into the start-up
ecosystem. Start-up companies’ expenses tend to exceed their revenues as they work on developing,
testing & marketing their idea and need funds to expand their business at different stages of
development. This project aims at analysing Indian start-ups from funding, failure and profitability
perspective and to predict the outcome of start-ups in India.
Keywords: Start-ups, Incubators, Legal Support, Fast-tracking, Tax Exemption, Funding, Profitability

2.1 Analysis of Factors Affecting the AQI of Delhi

New Delhi is among the most polluted cities in the world today. Air pollution is responsible for many
health problems in the urban areas. Of late, the air pollution status in Delhi has undergone many
changes in terms of the levels of pollutants and the control measures taken to reduce them. The
situation worsens during the onset of winters every year from October onwards. There are various
causes contributing to the current state of air pollution in Delhi.
There were conflicting reports on media on the actual cause of air pollution in New Delhi. Through
this study we hope to develop some insights that can help organizations (State/Central Pollution
Control Boards and NGOs) to advocate more stringent policy frame work to control air pollution.
There are multiple factors which affects the air quality, such as: stubble burning in NCR, construction
activities, vehicular movement, firecrackers, industrial pollution, diesel gensets, etc. But, the data for
all of these is not available readily. We could collate the metrological data from CPCB & an estimate
of vehicular traffic through google maps, & thus decided to restrict our study only to these two
sources of data.
An estimate of pollution caused by vehicular movement can be studied if the data on vehicles
running on road at any given point of time is available. However, Google Maps provide only data in
terms of time it takes to move from point A to point B. Since, the speed of vehicles on road majorly
depends on the density of vehicles, we decided to take speed of vehicles as an estimate of vehicular
density. The same was calculated using the time & distance we got from the google maps.

Keywords: Air Pollution, Metrological Data, Vehicular Movement

2.2 Predicting Graduate Engineers Employability

Graduate employability is an increasingly major concern for academic institutions and assessing student
employability provides a way of linking student skills and employer business requirements. In the last
four years, there is no significant improvement in employability of engineers. Recent study by Aspiring
Minds NRE Report shows that only 17.91% of engineers were employable for the software services
sector, 3.67% for software products and 40.57% for a non-functional role such as Business Process
Outsourcing. Student’s employability is a major concern for the institutions and predicting their
employability beforehand can help in taking timely actions in order to increase institutional placement
ratio. To know weakness before appearing for interview of any company can help students to work in
areas that they need to improve in order to best match the skillset required by company. Enhancing
student assessment methods for employability can improve their understanding about companies’ in
order to get suitable company for them. Data mining and predictive modelling technique such as
classification and regression is best suited for predicting the employability of students. The application of
data mining in student employability is to search for significant relationships such as patterns, association
and changes among variables in datasets. It provides classification methods to predict the level of
employability for students.

Keywords: Graduate Employability, Software Services, Business Process Outsourcing, Data Mining,
Predictive Modelling

2.3 Study of Family Planning Methods by Indian Women

SHOPS Plus is promoting the adoption of safer sexual practices, including delaying sexual debut and using
modern contraception to avoid unintended pregnancies. Our project is a subset of this project and aims
to determine patterns in the selection of family planning methods by Indian women using data analytics
techniques.
The groups obtained by analysing the data will reflect the choice of contraceptive given a woman's socio-
economic and demographic indicators. These group definitions can be used to promote relevant family
planning products and schemes by the government. We will also develop a predictive model using the
information obtained above. However, in absence of any new information to test the model, it may only
serve an academic purpose at best.

Keywords: Safer Sexual Practices, Data Analytics, Socio-Economic, Demographic, Indicators

2.4 Credit Risk Prediction, Customer Segmentation and Roll Rate Analysis

Credit risk profiling is very important for banks and other lending institutions. Profiling risky segments can
reveal useful information for credit risk management. They decide who is credit worthy and who is not
based on the individuals’/companies demographic information, credit history with the bank (if available)
and the information available at Credit Bureaus. Credit providers often collect a vast amount of information
on credit users. Information on credit users (or borrowers) often consists of dozens or even hundreds of
variables, involving both categorical and numerical data with noisy information.

This is a typical classification project where the company seeks to identify which clients are creditworthy.
The project also seeks to provide the following solutions to the company:

- Segment customers such that the company can develop marketing strategies to target specific
segments which will increase the potency of the marketing strategy
- Compute roll rates to predict credit losses based on delinquency

2.5 Automatic Summarization of News Articles

Text summarization can be done in two ways: Extractive text summarization, where the algorithm picks
important words from the article to generate a summary of the article, and
Abstractive text summarization where algorithm generates a summary of the articles with its own words
just like a human generated summary.
Automatic summarization of news articles was done using both the above ways and different sub
techniques with different results. Simple NLP based method using sentence scores was used to extract
top 5 sentences of the articles, it was observed that this method ignores the importance of nouns like
names, places etc., and hence the output may not be meaningful.
TD-IDF based method was also used to derive the sentence scores and then extract top 5 sentences for
the summary, results were impressive and all the nouns, places were given due importance however
there was still a drawback that this approach ignores the order of words. Lex Rank and text Rank
techniques were also used to summarize the text and results obtained had a good rouge score. The
abstractive text summarization techniques use deep learning models to generate a summary like a
human written summary. LSTM model was used to generate the same and the results obtained were not
as expected which were attributes to the fact that RNN model requires high configuration machines to
train the model which were not possible at this point of time considering the scope of the project.

Keywords: Text mining, abstractive text summarization, extractive text summarization, natural
language processing, tf-idf, Python
 2.6 Telecom Customer Segmentation and Churn Prediction

In this project we present analytical models for two very prominent use cases prevalent in telecom
industry i.e. Subscriber Segmentation & Churn prediction. The models thus created are based on proven
statistical frameworks and include domain expert inputs at relevant stages of the model to give practical
outputs and insights.
The analytical models are based on actual subscriber generated data points across various network and
business support systems of a telecom operator. We have used 24 weeks of data collected over ~4000
subscribers across various services such as Data, Voice (On-net, Off-net), Roaming, SMS etc. and where
required created relevant features from a combination of available variables. The data is mined using
statistical techniques and specifically use widely accepted and used analytical techniques i.e.
− Clustering (K-means algorithm)
− Predictive analysis using classification algorithm (Logistic Regression algorithm)
The models are cross-validated against equally acceptable analytical algorithms for benchmarking the
outputs. While creating the models in the first place, desired best practices of building right number of
samples as well as transforming data to conform to the statistical requirements of the algorithms have
been followed in letter and spirit to build robust models. Further, the models have been tested for their
stability using unseen data (in case of churn prediction) and validation techniques (in case of clustering).
As a result, we can present to business a clear view into four type of segments the cluster of subscribers
throws up. We also have labelled the overwhelming characteristics displayed by the members of a cluster
with respect to the type of services they consume whether Data, Voice, SMS, all or none (idle/dormant
users).

Keywords: Telecom Industry, Subscriber Segmentation, Churn Prediction, Clustering, Predictive Analysis

2.7 Predict Defaulters using Machine Learning

Credit Card and loan payment default is the one of the biggest issues faced by the retail finance industry.
It has a cascading effect as increasing defaults are compelling these organisations to contract their credit
card and loan portfolios immensely. The design of efficient fraud/default detection algorithms is key for
reducing these losses, and more and more algorithms rely on advanced machine learning techniques to
assist investigators. In this project, we have proposed the overall process of detecting the probable
defaulters based on machine learning and process large amounts of financial data. F-measure and ROC
curve are used to validate our proposed model. We also further cluster defaulters for further business
information.

Keywords: Machine Learning, Payment Defaults, Financial Defaulter Detection, Supervised Method,
Unsupervised Method, Feature Selection, SMOTE

2.8 Stock Options Trading using Machine Learning

Stock markets are unpredictable as the fluctuations in the prices over time depend on several factors.
This makes the investors full of doubts. However, if we look to the stock markets and behaviour of stocks
prices in more detailed manner, we can observe patterns that define the generic evolution of the stock.
Based on forecasted trends, we have to build our hypothesis for the analysis so that we will use
Predictive Machine Learning Classifiers (e.g. Regression) with stock data that spans over a long-enough
period of time, we can approximately predict the behaviour of the stocks and the evolution of their prices
over time though Time Series Forecasting. After all our main objective is to invest in different stocks to
get maximum profits for certain level of risks, for that we are creating a diversified portfolios of stocks
with least standard deviations to minimize risks try to quantify that risk factors. To quantify the risk
probability, we use Value at Risk (VaR) so that we will be able to analyse the downturn in the given
portfolio. Use of derivatives instruments like options (call and put) on a given stock help to hedge or
maximize our return.

Keywords: Stock Markets, Forecasted Trends, Regression, Time Series Forecasting

2.9 Upsell Lead Prediction of IndiaMART Clients

IndiaMART.com is a B2B Online Marketplace for Buyers and Sellers where Sellers showcase their products
and Buyers search / browse the platform to find Sellers for their requirement. The clients (Sellers) are
given Buy Leads through its Web enabled platform, enabling Buyers to find relevant Suppliers and vice-
versa. This project attempts to solve IndiaMART’s problem of identifying on-boarded Silver Clients on the
platform who are the most likely candidates for upsell to Gold / platinum package, thereby maximising
ROI for both client and IndiaMART. With the ready dataset, various algorithmic techniques such as
Logistic regression, Random Forest, Combined Probability – Neural Net and Neural Net Ensemble were
tried. The results were then presented to the business where business put forward further expectations
to be able to predict the exact service which must be pitched to client. Business also demanded to know
that gestation period for each client upsell.
Data with respect to the above requirement shall then be collected and further be presented to the
business. For now, the model is being sent for implementation where at the start of each month, we will
use this model to identify prospective clients which should be attempted upsell in the month.

Keywords: Online Marketplace, Logistic Regression, Combined Probability

3.1 Exploratory Study to Understand Driver’s Behaviour and Vehicle Condition

Driver’s behaviour strongly impacts traffic security and performance of the vehicle. Aspects of driving
such as over speeding, excessive engine idling, harsh braking has major impact on fuel consumption and
road safety. Telematics devices plugged in vehicle can play a beneficial role in promoting safe driving thus
reducing the accidents and their associated costs. Moreover, insurance companies are now trying to
adopt UBI (usage-based insurance program), which calculates the premium based on the driver’s driving
behaviour and time of driving because reckless driving badly affect the profitability of insurance
companies. Therefore, it is essential to have a system that can accurately profile a driver based on his/her
driving behaviour using both real time and long-term data, as well as other related vehicular and
environmental data. Telematics can also bring about benefits for tracking Vehicle health by the close
monitoring of different aspects of vehicle performance, including engine malfunctions, faults or warnings,
thus avoiding vehicle downtime. Thus, the problem this project tries to address can be defined as follows:
- How to profile a driver based on his/her driving behaviour using real time and long-term data?
- How to analyze whether vehicle needs maintenance or not?

Keywords: Traffic Security, Over Speeding, Harsh Braking, Telematics, Insurance Program

3.2 Cancer Diagnosis Transformation

Purpose of this report is solely for academic. Selection of healthcare as domain was based on considering
the significant contributions that can be provided by machine learning. The accelerating power of
machine learning in diagnosing disease and in sorting and classifying health data will empower physicians
and speed-up decision making in the clinic. Cancer is the second leading cause of death globally; early
detection is crucial and can go a long way in saving lives. We have tried to automate the manual
classification process done by the pathologist, wherein cancer variations are classified into mutation
classes. The key role of pathologists is to examine the report to determine precise type and
severity of the disease. Pathologists are integrating AI and machine learning software into their
workflows to cut down work time, reduce misdiagnoses and connect with fellow pathologists–in general,
making their departments more cost-effective and high functioning. Over the course of a cancer patient’s
diagnosis and treatment, pathologists record highly descriptive and specific observations of cells and
tissues in pathology reports. It is meant to serve as a tool to save pathologists’ time by instantly
highlighting potentially problematic areas of the scan, instead of having to do this manually. We have
focused to work towards enhancing the work of pathologists by building a model to classify the genetic
variation based on research literature into one of the 9 mutation classes. We have obtained dataset from
‘’Memorial Sloan Kettering Cancer Center (MSKCC)” website, which is one of the world’s oldest and
largest private cancer center that provides exceptional patient care and does lot of innovative research
on cancer. We have built 5 different classification models and 1 ensemble model and obtained
outputs for genetic variation classification into one of the 9 mutation classes with multiclass log-loss as
our key performance metric and created graphs in Python and Tableau for visual interpretation.

Keywords: Healthcare, Physicians, Cancer, Pathologists, Mutation, Python, Tableau

3.3 Sales Promotion Model for Big Bazaar

Retail chains world over spend a substantial part of their marketing budgets on promotions. While
promotions account for a significant share of retailer revenues, promotion intensity continues to
increase. Despite substantial investments in promotions, retailers often have little understanding of
the true performance and thus struggle to determine which promotions are working and which are
not. Future Retail runs various loyalty programs and festive offers which provide their customer more
opportunities to avail discounts. Customers can use these offers or loyalty program to either avail
discount or to make payments.
This project is aimed to analyse customer buying patterns using the data collected during the time
spanning 01/01/2016 to 06/30/2017. Eventually evaluation of performance of historical promotions
run and present recommendations on appropriate promotions to increase wallet share of customers
would be the successive step.
There are two ways to measure effectiveness:
1. Scale of efficiency—the extent to which cost of promotional spend is minimized for a given
number of promotions availed by customers
2. Scale of monetary improvement—the extent to which spend on promotion reaps/achieves
an increase in sales or profit
The project is aimed to evaluate effectiveness of promotions on monetary improvement and
thereafter to recommend improved promotional schemes to further improve the sales.

Keywords: Retail Chains, Retailer Revenue, Promotion Intensity, Loyalty Program

3.4 Credit Risk Analytics of Peer to Peer Lending Models

Credit risk is a serious problem in banking and is of concern to a variety of stakeholders -institutions,
consumers and regulators. It has been the subject of considerable research interest in banking and
finance communities and has recently drawn the attention of statistical researchers.
Credit risk is the risk that the borrower from a bank will default on the loan and/or the interest payable,
i.e. that it will not perform in terms of the conditions under which the loan was granted. This is damaging
to the bank, not only because of the actual loss eventually incurred, but also in terms of the time that
management and bank counsel expend on attempting to recover the loss or a portion of the loss.
The agenda of this project is to create multiple Credit Risk models using multiple techniques such as
regression and random forest try to develop and identify the best model that could predict the
probability of default with maximum accuracy and with least false positive rate. The model developed
through study would benefit the Banking companies to predict PD with higher degree of confidence level
and consequently would lower the value of NPA of the bank. With a healthy banking sector, and the loans
reaching to the right consumers the banking sector, the economy of the
nation as well as the consumer would prosper.

Keywords: Credit Risk, Regression, Random Forests, Probability, Banking Sector

3.5 Revenue Maximization in Retail Stores through Effective Campaign and Personalised
Recommender Systems

In this project, we explore the problem of purchase cart recommendation in the field of brick & mortar
retail stores. Recommender Systems application is widely used in online sales platforms and content
providers for e-commerce, movie recommendation etc. We want to explore usage of recommender
systems in brick & mortar stores while keeping in mind their operational constraints, specifically
restricted interactions between customers and products. Physical shops have to cope with new
competitors, e-commerce actors like Amazon, Flipkart & others, who have redefined ways customers
purchase - faster, cheaper and often more targeted. Large physical retail store chains have started to
compete on same ground, exploring online selling portals of their own. But, major chunk of their sales is
still from physical stores. We are proposing to use recommender systems to provide customized services
to their customers, thereby improving customer satisfaction, enhance their loyalty and thus increase
their benefits. One possible application of such a strategy would be to use personalized recommenders to
provide purchase cart recommendations, rather than individual product suggestions. We studied
different types of recommender systems and decided to implement a customized recommendation
engine strategy to achieve our goal. We expect our solution to bridge the gap between physical stores
and online platforms as far as customized customer experience is concerned and act as a valuable brick &
mortar retail store campaign strategy.

Keywords: Recommender Systems, E-Commerce, Customized Services

3.6 Anti-Fraud Bot for Bajaj Allianz Life Insurance

The main objective of this project is to develop a chatbot that can help the customers in answering the
queries instantaneously around the clock without having to wait for a human to assist them on their
queries. In the financial service industry, customer satisfaction is as equivalent or even more important
than to acquire a new customer. And also it is difficult for the internal employees to remember the
policies and procedures followed by the organisation. Our chatbot will be the answer to both the crisis.
With the help of deep learning, the chatbot is designed intelligently to cognizant itself from its past
experience and also redirect to the users to the correct point of contact if it fails to recognize the
question. That being said, more the questions asked, wiser will become the chatbot. Demand for chat
bots that imitate human conversations for solving various tasks is steadily increasing. Chatbots are the
new holy-grail of augmenting customer engagement and brand presence. With the help of intuitive
chatbots, insurance companies are able to explain complex products to their customers, drive brand
engagement, and improve sales and distribution with minimal human effort and intervention.

Keywords: Chatbot, Deep Learning, Customer Engagement, Brand Presence

4.1 Analysis of Claims Data for Invest Assure

Purpose – To provide a brief analysis of Invest Assure endowment policy claims management using
claims data of policy holders provided by Bajaj Allianz Life Insurance Company Limited (India).
Design/methodology/approach – For convenience, this analysis of claims data is somewhat arbitrarily
classified by topics as follows: age-based claims analysis, demographic analysis, smoking habit based
claims analysis, estimation of distribution of claims using six-year data, develop a model to screen the
future prospects (Customers) to approve the policy or not and prediction of future claims.
Findings – Emphasis is placed on a discussion of different kinds of characteristics affecting the claims on a
comparison of the statistical models and methods used to analyse such data. Prediction of future
customers who might claim is also performed using the model.
Project limitations/implications – Since the literature on claims management data is vast, more work on
this problem is needed. Due to data limitations, in-depth analysis on reasons of claim, connecting it with
mortality rate, and other such deep-dives weren’t feasible.
Practical implications – This report points out how claims management can be made more efficient. This
analyses can be performed and applied for all insurance types and across the insurance industry not
keeping it specifically limited to Bajaj.
Originality/value – The report reviews different statistical models and methods used to analyse claims
data. The statistical models and methods presented are valuable and meaningful for claims management
analysis and future claims prediction.

Keywords: Endowment Policy, Demographic Analysis, Claims Analysis, Statistical Models

4.2 Financial Analytics on Non-Performing Assets (NPA) in Banks of India

The topic of the study is “Financial Analytics on Non-Performing Assets [NPA] in Banks of India”. This
analytics engagement is carried out for the Banking and Financial Services Industry in India in general,
and not for any specific Company or Financial Institution. This study is conducted on the NPA of
commercial banks for the period 2007 -2017. This study helps the industry, non-banking financial
institutions to benchmark and the investors and businesses to understand the behaviour of NPA for
making right business decisions. The study analyses the NPA data and identify whether the NPA ratio is
same or different across commercial banks using ANOVA technique. The study also creates clusters using
K-means clustering algorithms and predicts the cluster group using Linear Discriminant Analysis. The final
step of the study forecasts NPA across commercial banks using multiple linear regression model.

Keywords: Financial Analytics, Non-Performing Assets, ANOVA Technique, K-means Clustering Algorithms,
Linear Discriminant Analysis

4.3 Predicting Trade Price of the Bond

Bond prices are a result of complex markets, decisions and polices, and predicting the future price are
difficult due to dependencies and complexities. Wide gap between the amount of relevant information
available to trading equities verses corresponding trading corporate bonds, makes the task challenging.
Predicting the bond price becomes even more challenging due to the lack of full disclosed information,
and accuracy is not the only consideration–in trading situations, time is of the essence. Hence machine
learning in the context of predicting bond price should be fast and accurate.
The goal of this project is to develop a global model using the techniques and algorithms of machine
learning and a set of data describing trade histories, intermediate calculations, and historical prices made
available by Benchmark Solutions, a bond trading firm, in order to accurately predict bond prices. Our aim
is also to paying more attention to trading dynamic, various features and structure of the bond.

Keywords: Bond Prices, Future Price, Machine Learning, In-Trading

4.4 Predicting Long Term US Treasury Yields

This study discusses the various factors which potentially influence the evolution of the long-term US
treasury yields. This has been an area of interest across the financial sector given the advent of number
of hedge funds, investment banks, mutual funds etc. who are very active in the US interest rates market
from a trading perspective. Additionally, long terms US yields are one of the key determinants of
mortgage rates in the US. US Mortgages and securitized products market is one of the biggest markets in
the world. Given the influence of these yields in the pricing of mortgages, the evolution of these yields is
not only interesting for speculators but also for hedgers who have exposure to these markets like
mortgage originators and servicers.
The study builds and compares various models using the variables which are assumed to potentially
impact the evolution of the US long terms yields and recommends the best model that can be used to
predict the direction and the extent of the future move in these yields. Lastly, the model also highlights
some of the challenges that the users should be aware of when using these models to make trading
decisions.

Keywords: Treasury Yields, Financial Sector, Hedge Funds, Investment Banks, Mutual Funds, Mortgages

5.1 RAG Classification for Measuring Performance of Life Insurance Agents

The agent/broker channel in the insurance sector plays a prominent role in the value chain. They are very
influential in determining the premium revenue, brand value, customer retention and customer service
experience. Insurers that can manage their agencies effectively are better positioned to achieve a
competitive advantage and hence the analysis of agency performance becomes crucial for an insurance
provider. The performance of the agents can be done on various parameters such as profitability, product
portfolio, customer retention as well as different servicing aspects of customer relationship management
such as complaints management, documentation, servicing customers based on requirement. Analytics
can help identify the good performing agents and provide them the necessary support to perform better.
It can also help in incentivising the good performers and identify good practices that can be shared across
other agents. Analytics can also help in understanding those agents that are consistently low performers
and either take corrective actions based on root cause analysis or prune such agents. At Bajaj Alliance Life
Insurance Company (BALIC), seven variables have been identified for performance measurement on the
customer servicing front. The performance of the agents is compared with pre-defined threshold values
for each variable and a final score is arrived at to classify the performance as Red, Amber or Green. BALIC
identifies the agents classified in the Red category and communicates areas of improvements that they
can work on to improve their performance.

Keywords: Insurance Sector, Premium Value, Brand Value, Customer Retention

5.2 Behaviour Scorecard

The organization provides credit to corporates. The objective of the project is to build a behaviour
scorecard to identify which of its live customers are likely to roll forward to higher delinquency bucket in
next few months, from which it is not possible to recover.
Two approaches were taken for data preparation. In the first approach, every 3rd transaction of the
customer was sampled for modelling. In the 2nd approach, a 3-month rolling performance window was
used to sample 1 transaction from each customer- thus maintaining the uniqueness of customer. For
each sampled transaction, historical data of previous 12 months was taken for observation of behaviour
(“observation window”) and next 4-6 months was taken to measure the performance (“performance
window”) of the customer. The resultant dataset was highly imbalanced, with just 3% (1.5% in 1st
approach) of the transactions representing the target class to be predicted. SMOTE technique, used to
bump up the minority class transactions, was optimized a 15% to give the best results. Logistic Regression
algorithm was selected as Business wanted to know the impact of the variables on the outcome. Due to
high multi-collinearity & large no. of features (72), model was run iteratively.

Keywords: Behaviour Scorecard, Delinquency Bucket, SMOTE, Logistic Regression

 5.3 Stock Market Analyst Rating

Analysts have been actively evaluating companies as long as there have been stocks, but they’re more
popular and get more exposure than ever thanks to round-the-clock stock market news and online
resources. Some analysts’ notoriety has also increased. But while analysts typically have similar
credentials, they aren’t all the same.
This project will help to differentiate among different stock analyst as per the recommendation given
by them. We have used the Machine Learning algorithm to create a framework which provides rating
to market analyst on the scale of 1 to 5 on the basis of their past performance.
Although a disclaimer that past performance is not a predictor of future performance accompanies this
promotional material, there is a strong implication that past performance is a good measure of the
market analyst’s skills.

Keywords: Stock Market, Shares, Financial Markets, Rating System, Predictive Analytics, Data exploration,
Data Cleansing, Classification, R, Excel, Random Forest, Naïve Bayes, Decision Tree, Machine Learning,
Market Capitalisation

6.1 Factors Influencing Infant Mortality Rate

This project is to identify the key factors that affect the Number of Infant deaths reported in India. The
Variable Importance is determined, and Infant Mortality is predicted by using below Machine Learning
techniques:
• Random Forest
• Linear Regression (using Principal Components)
• Neural Network
Apart from this, using the unsupervised learning algorithm (k means), to classify the data based on
the heterogeneity in the data points.

Keywords: Variable Importance, Infant Mortality, Machine Learning

6.2 Customer Segmentation and Profitability Analysis (for a Cash Management Company)

Cash Management and Logistics is an important segment of market in countries like South Africa.
Although Africa can provide great opportunities, a treasurer looking to run operations needs to be
cognizant of the risks and challenges associated with the region. We have partnered with one such cash
management company SBV Services Pty. Ltd. headquartered in Johannesburg, South Africa, to analyse
their customer segments and produce to them the key factors that are affecting their profitability within
each customer segment.
One of the important business problems of SBV was that they had various types of customers who have
subscribed to different types of services from SBV, however, the firm was not able to understand which
customer segment is profitable and the factors affecting the other customer segment that less profitable.
We collected the data pertaining to the customer details, customer type details, and transaction type,
vehicle run time, revenue, direct and indirect costs which will determine the profitability of the
transaction. The data was then subjected to a K-means clustering technique after obtaining the optimal
number of clusters through an elbow curve. The outcome of the analysis was that the company had 4
distinct customer segments of which two clusters were profitable and the other two were making loses.
The important factors that were affecting the profitability are service time of each cluster, indirect cost,
direct cost spent towards the staff which were dependent on the type of the customer viz. wholesale or
retail.

Keywords: Cash Management, Logistics, Customer Segments, K-means Clustering Technique

 7.1 Sentiment Analytics for Stock Market, Crypto, FX and Commodity

The purpose of this work is to observe how well the changes in prices of financial product, the rise and
falls, are correlated with the public opinions being expressed in tweets about the respective underlying
product. Understanding author’s opinion from a piece of text is the objective of sentiment analysis.
In this project, we would like to use applied sentiment analysis and supervised machine learning
principles to the tweets extracted from Twitter and analyse the relation between respective market
movements (viz. Stock Indexes, FX, Commodity & Crypto) and sentiments in tweets.
At the end we will try to find out whether there is a strong correlation exists between the rises and falls in
financial product prices with the public sentiments in tweets.

Keywords: Sentimental analytics, Text mining, Sentiments, Machine Learning, Twitter, Sampling, Trend
Analytics, Tidy Text

7.2 Predicting Cervical Cancer Likelihood and Indicators using Machine Learning

Cervical cancer is a type of cancer that occurs in the cells of the cervix. Various strains of the human
papillomavirus (HPV), a sexually transmitted infection, play a role in causing most cervical cancer. One
can reduce your risk of developing cervical cancer by having screening tests and receiving a vaccine that
protects against HPV infection.
To identify and confirm the presence of cancerous tissues in the body, various screening tests are
conducted. Screening means checking one’s body for cancer before one has symptoms. As in many other
diseases, the existence of several screening and diagnosis methods creates a complex ecosystem from a
Computer Aided Diagnosis (CAD) system point of view. Hence, there is need to predict individual patient’s
risk and the best screening plan during her diagnosis.
The objective of this project is to use Computer aided diagnosis and machine learning techniques to
classify respondents into Cervical Cancer or healthy category based on multiple test results provided in
the dataset thereby preventing the need of doing multiple screening tests. As the data is related to
cancer, our project team would need to have working knowledge of the relationship between cancer
genes and cancer progression. As 2 of our team members are in the pharma industry, we have access to
microbiologists and researchers to help us here.
The primary objective of this project is to avoid the above uncomfortable and painful situations by
predicting risk of cancer using historical data and do early detection without the need of doing multiple
screening tests. This reduces the burden of healthcare cost and patient compliance. The application of
the model is for both patients and clinics. For patients, it would be early prediction of cancer without
undergoing any painful screening tests. Only high-risk patients must undergo for screening tests for
confirmation. Early diagnosis (Stage 1 or Stage 2) will have more therapeutic options for patients with
much better outcomes. Our approach is to consume the data collated from surveys on the lifestyle
habits, medical history etc. from available sources conduct analysis using machine learning techniques
and conclude with a model that
• Helps to predict the likelihood of cancer without doing the different types of painful tests
• Identifying the parameters that increases the likelihood of cervical cancer

Keywords: Cervical Cancer, HPV, Computer Aided Diagnosis, Machine Learning, Pharma, Healthcare Cost,
Patient Compliance

7.3 Football Player’s Marketability Index

Football is the most popular sport in the world. It is played by almost 211 countries. Its popularity has
increased even more by the rise of club football. Clubs like Real Madrid, Barcelona, Manchester United,
AC Milan, Juventus, Arsenal, Bayern Munich and many more have huge fan bases. Teams continue to
form part of the cultural and sentimental heritage of cities and national teams, which continue to arouse
passions. They can be considered another national symbol. The fan base is so huge that it has gone way
beyond just the locals supporting the club to people from all across the world supporting it. With the
advent of Broadcast TV and Social Media, the game has become even more global. Football has changed
and so has its economics and marketing. Now, clubs buy big money players for two reasons. One is
obviously to make the club better by having someone with specific skill set and the other reason is to
help the club in marketing which in turn helps them in increasing their revenue, which helps them in
expanding their base to different countries. Paris Saint-Germain (French Club) bought Neymar from
Barcelona on a record shattering transfer by triggering the release clause of $263 million. On the back of
his transfer, there was plenty of focus on shirt sales. Around 120,000 Neymar PSG jerseys were sold in
the 30 days that followed his switch—raising $10.1 million. Commercial value may be hard to judge, but
one thing that is more easy to monitor is social media numbers, which can measure how influence and
business is growing. Across Instagram, Twitter and Facebook, PSG’s followers at the end of the 2016/17
season totalled 38.1 million. That figure soared by 5.3 million ahead of Neymar signing his new contract.
Now, 12 months on from the transfer, it is apparent from the figures on PSG’s accounts that followers
have soared by a further 11.8 million across the three social media platforms. Take into consideration
also that Neymar has a personal following of 165 million and his reach to fans, businesses and consumers
across the world becomes hugely significant. Even in the Indian Super League, when the league started,
every team signed a marquee player. Players such as Luis Garcia, Alessandro Del Piero, Freddie Ljungberg
and David James were signed by different clubs to attract the Indian audience towards the league by
marketing all these players before the start of the season. After the season ended, clubs were asked to
retain 2-3 players for the next season. Kerala Blasters retained CK Vineeth and Sandesh Jhingan. Why
these two? Sandesh Jhingan was retained for being a good defender in the previous season but Vineeth
was retained for being the local boy which helped Kerala in their marketing.

Keywords: Machine Learning/Deep Learning, Sentiment Analysis, Player Rating System

8.1 Loan Approval Prediction based on Parametric, Non-Parametric and Tensor Flow Machine
Learning Models

Due to easy availability of loans, in today’s world, taking loans from financial institutions has become a
very common phenomenon. Everyday a large number of people make application for loans, for a variety
of purposes. But all these applicants are not reliable, and everyone cannot be approved. In the past as
well as this year, we read about a number of cases where people do not repay bulk of the loan amount to
the banks due to which they suffer huge losses. The risk associated with making a decision on loan
approval is immense. Today’s era is all about “Data Management”. Nowadays, banks have included a
large amount of information in its evaluation of loan issuance, and some of this information has a vague
causal relationship with the loan default rate. The growing amount of data due to improved data capture
and data storage technology has bring us to a new perspective on this problem which is used to be
accomplished by financial and economic analysis. Accurate prediction of whether an individual will
default on his or her loan, and how much loss it will incur has a practical importance for banks’ risk
management is very important. So, the idea of this project is to:
• Study to whom the loan was granted before and on the basis of these records/experiences and extract
patterns, which would help in predicting the likely defaulters, using classification data mining algorithms.
• Banking Big Data are stored in form of clusters, so understand the distributed environment and suggest
a suitable ML algorithm to work in distributed scenario.

Keywords: Loans, Financial Institutions, Risk, Data Management, Economic Analysis, Data Mining
Algorithms, Machine Learning
 8.2 Develop the Credit Scorecard for Lending Club

As a part of project, we studied the problem faced by the P2P lending industry, where lender can earn
higher rate of interest, whereas borrower can spend lower amount of interest as compared to the
conventional banking system, where the concern is how risky is the borrower & given the borrower risks,
should we lend him/her or not. After the thorough study we conducted, we were able to suggest a credit
scorecard for the Lending Club Corp., & explored various machine learning classification model
techniques for predicting the default rate of customers like Logistic Regression, Random Forest, Lasso &
Ridge & Ensemble & cross validated our data on the test data set.
Following deliverables will be met to address the problem faced by the Lending Club-

1. Analysis the loan payment dataset of the Lending Club Corp, to better understand the best borrower’s
profile for investors.
2. Perform segmentation of the loan database into finish cases & current outstanding loans.
3. Breakdown the composition of the default cases.
4. Examine the correlation between default cases & the independent variables/indicators.

Finally, the probability of the customer defaulting on the loan pay out was calculated for each individual
customer and then based on that the loans were classified into the likelihood of high, medium and low
risk of delinquency. The low risk customers should be given rebates in the interest rates which can
improve customer retention rate. The Machine Learning Algorithm or an analytical model can be further
extended to analyse the data sets available from all P2P lending companies available in the Market.

Keywords: Machine Learning, Logistic Regression, Random Forest, Segmentation, Correlation, Probability,
Likelihood

8.3 Accent Analytics

Accent is a distinctive way of pronouncing a language, especially one associated with a particular country,
area, or social class. English is very close to becoming a global language with an estimated 1.5 billion
speakers globally. Yet, it remains a fascinating subject because it is spoken in different ways across
countries. Among the many issues faced by Automatic Speech Recognition (ASR) systems, effectively
handling accents are one of the most challenging. Particularly when working with languages that have
highly varied pronunciations such as English, an ASR system trained on only one accent might only be
effective for a minority of the speakers of that language. The Objective of this project is identifying and
accurately labelling English accent from different geographical regions. There are many accents given the
widespread use of the English language. Therefore, we limited ourselves to the four accents: American
English, Arabic, Mandarin and Indian. The main objective is to design and train a model to identify and
label these four accents in spoken English.
Primary Objective of the study are:
• The approach to this task of accent classification consists of feature extraction and developing machine
learning classifiers
• Extract acoustic feature MFCC (Mel Frequency Cepstral Coefficient) as an input to machine learning
algorithms
• Build various Machine Learning multi classification algorithms to find the best fit model
• Predicting / Classifying short 30-second accented voice clips as one of target Accents
The Audio recordings of people with different Accents are scrapped from the Speech Accent Archive an
online repository of spoken English. Also, Speakers representing different native languages read a
common paragraph in English was recorded.

Keywords: Accent, Geographical Regions, Classification Algorithms, Machine Learning, Predicting

 8.4 Product Analytics for Retail Brand Apparel_Alhassina

The projects would contain analysing the Brands & Apparel_Alhassina dataset and implementing product
analytics. The models have been built after doing missing value treatment, outlier treatment, EDA,
Univariate and Bivariate Analysis on the data. The 3 main objectives that we have work towards in this
project are as listed below-
Sales Revenue Prediction: A sales forecast is an essential tool for managing a business of any size. It is a
month-by-month forecast of the level of sales you expect to achieve.
Model Used – Linear Regression Model and Accuracy Check Commonality of product: Based on customer
characteristic which product is commonly purchased need to pitch the product to the customer based on
customers purchasing habits.
Model Used – Market Basket Analysis and Association Rules Customer segmentation: Identify customers
who are high/low spending customers and who are engaged with you and find which market has highest
consumers.

Keywords: Product Analytics, Missing Value, Outlier, Univariate and Bivariate Analysis, Sales Revenue
Prediction, Linear Regression, Market Basket Analysis

9.1 Travel Time Prediction and Cab Requirement Forecast for a leading Transport as a Service
Organisation

The main objectives of this project are:

• Predict and benchmark the travel time from different zones to MNC office and vice versa in order
to manage expectations of the employees using a regression model
• Forecast a range of weekly Cab requirement for login and log out shifts per zone and per vehicle
type (4 seater/6 seater) on the basis of historic data using a time series model

Keywords: Regression Model, Forecast, Historic Data, Time Series Model

9.2 Hybrid Model for Solar Power Forecasting

Integration of solar energy into electric network is one of the most promising approach to meet the ever-
increasing demand for energy, without significantly impacting the environment. To ensure efficient
operation of power generation and distribution systems, it is essential to have reliable forecast
information of solar resources. Forecasting enables quantification and estimation of the available energy,
and in-turn resulting into optimal management of transition between intermittent and conventional
energies. In this study, an attempt has been made to explore and evaluate different approaches for
forecasting the solar power generation for the next day. The approach developed in this study is based on
the simulations from Numerical Weather Prediction models (NWP). NWP uses mathematical models of
the atmosphere and oceans, to predict the weather using information about current and past weather
conditions. Results of these simulations are used as inputs for statistical model, estimation of bias and its
corrections, and finally to create models for forecasting the power generated from a solar power plant.
This very short-term forecasting service is very important for grid operators, it ensures the stability of the
grid, and aid the power plant operators to keep the plant deterministically controllable. Several different
forecasting techniques were investigated, and the most appropriate method for forecasting the daily
solar irradiance and the amount of energy produced by the plant was selected. Bhadla Solar Park,
Rajasthan, India, has been considered for evaluation of the prediction model presented in this case study.

Keywords: Forecasting Solar Power Output, Simulating Weather Parameters, Bias Correction, Grid
Optimization, Solar Power Integration
 9.3 Youtube Video Analysis

Analysis of structured data has seen tremendous success in the past. However, analysis of large- scale
unstructured data in the form of video format remains a challenging area. YouTube, a Google company,
has over a billion users and generates billions of views. Since YouTube data is getting created in a very
huge amount and with an equally great speed, there is a huge demand to store, process and carefully
study this large amount of data to make it usable. Today, YouTube is the largest user-driven video
content provider in the world; it has become a major platform for disseminating multimedia information.
A major contribution to its success comes from the user-to-user social experience that differentiates it
from traditional content broadcasters. This work examines the social network aspect of YouTube by
measuring the full- scale YouTube views graph, comment graph, and video type corpus. There are a lot of
companies that are actively creating content on YouTube aiming to reach their user bases. The content is
posted in their respective channels. Content is either served directly on the channel or by pushing it on
other popular videos as advertisements. Companies have varying strategies to make their content
popular; some go the organic route while other are big on advertisements. The report documents the
effort to a data driven approach for understanding the YouTube market by taking client data of 35
channels. Through the historical video content, efforts will be made by applying different algorithms:
Neural Network, Logistic Regression and Random Forest to understand the accuracy of the algorithms in
predicting how much of a channel content gets hits through Advertisements. Time-series analysis is
performed to study the relationship between key variables and their variability. Finally, Multilevel
Regression analysis will be performed to understand the effect of these parameters and accuracy these
parameters would lead to. All the analysis and the findings will be delivered through statistical software
R.

Keywords: Multimedia, Comment Graph, Video Type Corpus, Neural Network, Logistic Regression,
Random Forest, Time Series, Regression Analysis

9.4 Bank Marketing

Recently, economic depression, which scoured all over the world, affects business organizations and
banking sectors. Such economic pose causes a severe attrition for banks and customer retention
becomes impossible. Accordingly, marketing managers are in need to increase marketing campaigns,
whereas organizations evade both expenses and business expansion. In order to solve such problems,
data mining and analytics techniques are used as a factor in data analysis, data summarizations, hidden
pattern discovery, and data interpretation. In this project, various data analytics techniques have been
implemented, using a real marketing data obtained from Portuguese marketing campaign related to
bank’s term deposit subscription. The project aims to improve the efficiency of the marketing campaigns
and helping the decision makers by reducing the number of features, that describes the dataset and
spotting on the most significant ones and predict the deposit customer retention criteria based on
potential predictive rules.

Keywords: Economic Depression, Banking Sectors, Marketing Campaigns, Data Mining, Hidden Pattern
Discovery, Customer Retention

9.5 Out brain Click Prediction

The internet is a stimulating treasure trove of possibilities. Every day we stumble on news stories relevant
to our communities or experience the serendipity of finding an article covering our next travel
destination. Outbrain, the web’s leading content discovery platform, delivers these moments while we
surf our favourite sites. Currently, Outbrain pairs relevant content with curious readers in about 250
billion personalized recommendations every month across many thousands of sites.
In this Project, our objective is to predict which pieces of content their global bases of users are likely to
click on.

Keywords: Travel Destination, Content Discovery Platform, Prediction, Global Basis

10.1 Consumer Loan Analytics

The key objective of the project was identifying variables and consumer segments that will allow lending
to larger audiences and simplify the current lending process.
This was a challenging project at multiple levels given the big data dataset spread across multiple
variables and no access to client for clarification/ reasoning or professional infrastructure. However,
overcoming the challenges, team has extracted a data subset 2 which has been used for this project. Data
modelling has been first done by using Logistic Regression and later by Random Forest technique. These
techniques were identified as best suited techniques given the nature of dataset (independent,
dependent variables and binary output) and the business objective being exploratory in nature. Multiple
tools like R, Python, and Cloud were used for data merging and modelling. With Logistic Regression team
was able to identify 55 significant variables mainly spread across Demographics and prior Lending history
like amount, credit period, future instalments, etc. This model had accuracy of 90.5%. These findings can
be further refined using neural network and XGboost model. It is possible that there are certain findings
that might come across as generic in nature; the final recommendations need to be discussed and agreed
with the business team for implementation. Given the generic nature of findings, they can be
implemented by financial companies across board and not just ‘Home Credit’. Business teams need to
understand the geographical nuances and legislative restrictions before implementation of the findings.

Keywords: Customer Segments, Data Modelling, Logistic Regression, Random Forest, Neural Network

10.2 Health Economics and Outcome Research of a Public Health Insurance Organization

While India is passing through demographic and environmental transition, the landscape of healthcare is
witnessing constant addition of disease burden, putting strain on individual health as well as public health
resources. Total healthcare expenditure in India is estimated to be Rs. 5.3T, amounting to 3.8% of
country’s GDP. Government alone spends nearly Rs. 1.6T on several healthcare initiatives. Targeted
healthcare can help reduce this burden, but administrators at healthcare organizations need to develop
data-driven strategies to achieve this. In the current study 5-year data of a particular public health
insurance organization is analysed to provide insights on patient attendance in the clinics, overall
expenses of the Organization, patient profiling on the basis of prescriptions & treatments, hospitalization
claim pattern and resulting financial burden. Patient attendance patterns were studied in the clinics
between 2014 and 2018, and short term forecast for next 6 months ahead was generated using time
series forecasting. It is observed that the increasing attendance in the clinics has led to 70% increase in
the median waiting-time of the patients over the past 5 years. G/G/m Queueing model was constructed
to estimate patients’ waiting-time in three clinics. Suggestions are made to improve the waiting-time of
the patients by reducing non-clinical activities and optimizing number of doctors in each clinic. Reduction
in waiting-time can significantly improve patients’ satisfaction, while increased availability of doctors
would improve the quality of diagnosis and prescriptions. Expense data analysis showed increasing trend
in expenses under most categories. Category-wise expense forecast was generated for next two financial
years, which estimated total expenses to increase from Rs. 1.07B in FY 2018-19 to Rs. 1.23B in FY 2019-20
and Rs. 1.35B in FY 2020- 21. This would help the Organization plan and propose budget for coming
cycles. Patient profiling based on prescribed medicines and hospital treatments showed significant
increase in certain clinical conditions. Anti-diabetic drug prescriptions have doubled, and emergency
hospitalizations due to Respiratory ailments have increased 90% over the past 5 years. This information
can be used for targeted/preventive care and planning special programs by the Organization to improve
beneficiaries’ health. Financial implications of the Organization’s medical expenses on the public funding
was studied, and Insurance Claim Ratio (ICR) was calculated. The study showed ICR of 637%, which is 6 to
10 times higher than any private or public sector health schemes operating in India. Recommendations
are made on limiting the scheme’s ICR, which in turn will reduce burden on public funding.

Keywords: Healthcare, Prescriptions, Treatments, Claim Pattern, Time Series Forecasting, Preventive Care

10.3 Churn Management

All industries suffer from churn – the loss of customers to the competitor. The survival of any business is
based on its ability to retain customers. This is particularly true for Telecom Companies where cost of
customer acquisition and retention is very high. The advanced analytical techniques can help CSPs reduce
the Churn by proactively identifying potential churners and taking timely retention measures.

Keywords: Customers, Competitor, Retention, Telecom Companies, Customer Acquisition, Analytical

Techniques, Churn

11.1 Sentiment Analysis for Customer Service in Banks

The purpose of this report is to is to improve customer experience by utilizing AI and machine learning to
develop an algorithm where customer E-mails are scanned, analyse the sentiment from the body of the
message and automate customer e-mail categorization and prioritization for the banking sector. The
main goals are stated below:
(1) Collect bank query related E-mail data, ranging from general information, escalations and requests
(2) Develop a machine learning algorithm that can perform text mining and sentimental analysis
(3) Provide priority based categorized information for management to prioritize and improve customer
Service

Keywords: Customer Experience, Machine Learning Algorithms, Automation, Banking Sector, Text Mining

11.2 Patent Based Technology Landscape Analysis and Grant Prediction for Trading Strategy

Investing in the right company has always been a challenging problem for investors. While large financial
data sets and indicators are available, there may be implicit value drivers that are not being considered.
One of the key value drivers especially for product based companies is the patent portfolio. In our view, a
deeper analysis of the patent portfolio and its value will provide investors with an excellent tool to
understand the growth of the company and drive investment decision making. Using a few large
electronics companies, we compare how the business performance in terms of revenue and stock price is
impacted by the patent portfolio of the companies as well as R&D expenses. We find that companies that
have deep investments in patents like Samsung and Apple have a statistically significant correlation
between revenue and stock price to the number of patents filed and granted. Given that patents are an
important indicator of business performance; we proceed to formulate a trading strategy that is based on
a forecast of the number of patents applied by the companies. Given the large number of patents being
filed and the high cost associated with the same, companies need an analytics based tool to decide on
patentability. Using machine learning models and NLP techniques, we predict whether a given patent will
be granted or not. This will help the companies use the model as a dipstick before applying for patents.
Given the wealth of information available in patents, an important opportunity for company leadership is
to understand how the technology landscape is changing in the industry as well as for competitors. We
use NLP based analysis on patent abstracts to determine how the technology landscape is changing over
the years and where fresh investments are happening now. The purpose of all of the above patent
analytics is to provide enabling tools for both the investor and an interested company’s leadership.
Keywords: Patent Analytics, Trading Strategy, Grant Prediction, Hypothesis testing, Technology
landscape, Machine Learning, Natural Language Processing, Latent Dirichlet Allocation, Legal Analytics,
Time Series Forecasting, Word Cloud

11.3 Analysis of Retail Store Sales using Machine Learning

A major challenge for large retailers is to address the needs of the consumers more effectively on a local
level, while maintaining the efficiencies of central distribution. As the demand for mass customization by
consumers grows, methods focused on store level optimization increase in value. The recent
development in information technology and ubiquitous computing makes it feasible to move
recommender systems from the ecommerce realm, where they are widely used, to retail stores. The
world is moving towards online shopping due to the benefits it provides to the customers. Big e-
commerce sites like Amazon, Flipkart etc. provide tailored recommendations to their customers based on
their previous behaviour. These aid the customers during their shopping and provide satisfactory
shopping experience. These recommendations are given by recommender systems employed by the
sites. As these systems evolve and become more complex, more accurate predictions and
recommendations are given to the users. If this continues, retail business might wane in the future.
Hence, a system which uses cash receipt data to provide personalized recommendations to customers
can help provide business advantage to the retail stores.

Keywords: Retailers, Central Distribution, Information Technology, Accurate Predictions

11.4 Financial Risk Modelling for Corporate using Machine Learning

The credit risk defaults by companies have been increasing at an alarming rate which may cause huge
unexpected monetary loss to the banks or financial institutions. This may directly affect the functionality
of the financial institution to a great extent. In the above context, the analysis of the risk profiles of the
companies becomes more significant. Banks or any financial institution should be able to predict the
companies which are likely to default through the risk profile analysis and avoid lending to those
companies. The primary objective of this project is to analyse the past and current financial data and
predict the future risks. The process of analysis deals with categorizing the companies into one of the
three risk profiles namely, low, medium and high. The project uses the following 6 machine learning
algorithms to predict the risk profile of the companies.

1. Support Vector Machine

2. Linear Discriminant analysis
3. CART
4. Random forest
5. K-Nearest Neighbour
6. Naïve Bayes

The project uses the following parameters to assess the performance of the models.

1. Accuracy
2. Average Sensitivity
3. Average Specificity
4. AUC

Keywords: Credit Risk, Financial Data, Machine Learning Algorithms

 11.5 Exchange Rate Forecasting Modelling

Forecasting the exchange rates is a challenging and important task for the modern day traders, people
associated with the financial markets and general population across the globe. In this report we will be
utilizing the time series concepts to do an analysis and predict the daily exchange rates of the Indian
Rupee (INR) against the United States Dollar (USD). We have investigated and compared different
techniques like using EMD, ACF and PACF, Neural networks, Support Vector Regression and Additive
Regression. We have used the daily exchange rates of the Indian Rupee (INR) against the United States
Dollar (USD). The foreign exchange market is the largest financial market in the world and forecasting
exchange rates are not solely an important task for investors, but also for policy makers. Since market
participant do not have access to future information, they try to model the exchange rate by past
information. Forecasting foreign exchange rate is one work that supports to foreign exchange rate risk of
commercial joint stock banks. This paper explores the behaviour of daily exchange rates of the Indian
Rupee (INR) against the United States Dollar (USD) and forecasts the forex rates for next 365 days. We
use statistical models ARIMA, and Neural Network to examine the performance and in forecasting the
currencies traded in Indian foreign exchange markets. Daily RBI reference exchange rates from January
1960 – May 2019 were used for the analysis.

Keywords: Time series, Machine Learning, Linear regression, Classification & Regression Trees, Random
Forest, Support Vector Machines, K-FOLD cross validation, KNN, Neural Networks

11.6 Customer Demographic Clustering and Pricing for Health Insurance Plans

An existing life insurance company is launching a new set of health insurance plans in India and would like
to do a pre- launch analysis on the market base and identify the optimal price range for different types of
customers to better aim the marketing pitch at the target audience with attractive pricing option at the
first glance. This is not an easy feat considering how insurance products can have varying range of
premium amounts and different set of benefits specific to the requirements of customers. Therefore, a
sample customer base via survey was collected and the data was used to cluster the market into varying
segments and zones to facilitate the marketing and quick pricing. The survey data contains attributes
related to the customer demographics, their illnesses or family illness history, physical disability and
information on existing insurance plans they may hold. The metrics are then cleaned up and clustered
using both k-modes and k-means, to facilitate comparison and it is observed that k-modes returns better
results in case of a purely categorical dataset. The survey data yielded 8 distinct clusters and it was
observed that almost all the risky customers were picked out to belong in a single cluster which will be of
utmost business importance. The model can be expanded to apply across a large customer base as well,
and to provide quick premium quotations or to design marketing campaigns.

Keywords: Clustering, K-Modes Clustering, Customer Demographics Clustering, clustering in Python, K-

modes in python, Clustering in insurance

11.7 Prediction of Thyroid Condition

This project involves the Prediction of the Thyroid Condition in the Domain of Life sciences. Life sciences
discoveries are helpful in improving the quality and standard of life, and have applications in health,
agriculture, medicine, and the pharmaceutical and food science industries. This analysis is conducted by
using machine learning techniques like CART and logistic regression classification algorithms to predict
the condition of thyroid in patients. The project also provides insights on Thyroid diagnosis and detection
depending on the given variables. In this project, we are focusing on the prediction of Thyroid condition
on potential Humans so that we can help in preventing this from occurring.
Keywords: Prediction, Machine Learning, Classification and Regression Trees, R, Logistic regression,
Recursive partitioning (Tree based) models – CART, Hypothyroidism and Hyperthyroidism

12.1 Predict World Cup 2019 Winner by Calculating Team Strength

The primary objective of this project is to explore the possibility of developing a Predictive Model to
effectively predict the outcome of a match played in World Cup 2019. We aim to build a model that will
provide the below statistics:
• Player Performance Score (Batting & Bowling) - Identify the batting and bowling metrics that
provide insight into the performance of a Player – when evaluated as a Batsman and a Bowler
• Team Performance Score (Batting & Bowling) - Derive a methodology to calculate ‘Team
Performance Score’ (separate scores for Batting & Bowling) based on the individual Player’s
scores.
• Team Performance Score - Identify a method to arrive at a single ‘Performance Score’ for every
team that will take into consideration - the above batting & bowling performance scores, Fielding
efficiency & ICC Rankings
• Winning-ness Probability – Predict the chances of winning-ness for both teams in a match.
• Predict the final winner of the tournament

Keywords: Predictive Model, Metrics, Probability

12.2 Drug Store Analytics

One of the major impediments of successfully running a drug store comes from insufficient stocking of
medicines based on the demand in the store. In this study we have tried to explore the maximum
number of drugs available in the market in consideration to its targeted and measured effectiveness to
an array of ailments. Our problem statement encircles the issue faced by store management to have all
the medicines available in their store in their limited space. Along with this we have tried to focus on how
the store should not lose customers because of unavailability of medicines. The chance that a customer
visits the store again will increase if the store can sell all the asked medicines in their visit. In order to
start a drug store and run it successfully, the store needs to be started with sufficient medicines required
to run the store. However, major problem the store management will face is, it can’t have all the
medicines available in market inside the store due to limited space. At the same time the store should
not lose customers just because of unavailability of medicines asked. So, the drug store needs to keep the
most asked drugs with them. The chance that a customer visits the store again will increase if the store
can sell all the asked medicines in their visit. In case of chain pharmacies, they have other branches
where already business is going on. So, they can use that database for analysis to understand which
medicines are frequently sold. In case of hospital pharmacies, the doctors can tell what drugs need to be
maintained at store as they are the ones who will prescribe medicines to the patients who visit their
hospital. However, in case of independent stores they don’t have any database which is already present
or any doctor who is dedicated to the store. So, identifying the frequently bought medicines is a bit of
challenge in this case. So, in order to identify the most asked medicines for an independent pharmacy we
have downloaded data that contains online reviews given by patients who used the medicines. More the
number of reviews, more the medicine is used. Better review ratings indicate that the drug is effective
and did not cause any bad effects on users. The data is studied and a list of medicines which are used
more frequently has been created. Excel, Tableau and R are used for cleaning, exploring the data and
building model. The store managers should keep these medicines in the created list at store in order to
avoid stock out scenarios and provide customers the medicines asked by them. This improves the
customer satisfaction and results in repeat business and loyalty as the customer will be happy if they can
get all medicines in one place without roaming around different shops for some of the medicines in the
prescribed list.
Keywords: Insufficient Stocking, Store Management, Chain Pharmacies, Model Building

12.3 Yelp Data Analysis

Everyday millions of people visit hundreds of restaurants in US and review them based on their
experiences at Yelp. Yelp wants to use this information and classify restaurants into good restaurants and
bad restaurants and recommend its users the best possible restaurant in their locality. On the other side,
it can also offer business recommendation and consultancy as a service to the local restaurants based on
the category in which a particular restaurant falls into.
Yelp has business related information like – business id, business name, ratings, demographic information
in business json file and review related information in review json file which has information of the
review of a particular business, star ratings and other attributes. Using this information our goal is to
understand the sentiment of the reviews and get key insights of the reviews and build machine learning
models using Rstudio and python which can classify a particular restaurant into good restaurant and bad
restaurant based on the review text and the stars that are provided.
Once the restaurants are classified, yelp can use this information in two ways – Come up with
Recommendation to its yelp users and help them choose a better restaurant. The other way is to reach
out to the restaurants give them yelp’s categorization/rating and offer them consultation as service
which will help local businesses to bring up their game.

Keywords: Restaurants, Classification, Consultancy, Demographic, Machine Learning Models, Rating

13.1 Brand Logo Classification

As of January 2019, over 95 million photographs are uploaded on Instagram each day. Social media brings
with it a plethora of opportunities for marketers in terms of the sheer volume of images that have the
possibility of being uploaded to numerous platforms. Without the availability of tools to sift through all
these images to detect their logos in the images, brands will be unaware of the cascade of threats and
opportunities directed at them each day. Social media websites contain large amounts of un-labelled
image data, from which we can extract many useful insights (such as the share of voice of a digital event
and negative brand mention). Vigilant companies closely monitor their web presence by keeping track of
any mentions of their brand name on Twitter, Facebook, Instagram and other social media sites.
However, a large quantum of social media content consists of user- generated images. We understand
that globally, people share almost three billion images on various social media feeds, of which almost
80% of images do not have any textual mention to brands. Instead of tweeting the word ’KFC’, a user may
instead post an Instagram photo containing a KFC merchandise. The company in question, or third-party
analyst, could find detecting such an event beneficial. There is no accompanying text mentioning their
brand name. The challenge is that their usual text analytic tools used for tracking the brand will be
completely stumped. We fathom that the ability to effectively detect such ’visual mentions’ will have
many commercial applications soon. The need for logo detection technology allows scanning images for
logos to get real uses of products by customers, facilitate monitoring the ROI of marketing campaigns,
ensure revenue boost, and more. There are few tools like Brandwatch Image Insights, IBM Image
Detection, Amazon Rekognition in the market that have identified this gap. These tools have been using
techniques such as Tensor Flow (Machine Learning), Deep learning algorithms to help analyse these
images. Researchers have developed various methods for decision making using supervised and
unsupervised classifiers. Recently, there has been increased attention to ensemble learning –methods
that generate many classifiers and aggregate their results. In this project, we explore a small component
of image analytics – classification of greyscale brand logos and building a model to identify the accuracy
and sensitivity of classification using advanced analytics techniques such as SVM, Random Forest, GBM
on R platform.

Keywords: Instagram, Social Media, Image Data, Insights, User-Generated Images, ROI, Machine Learning
 13.2 Assisted Demand Planning using Machine Learning for CPG and Retail

The evolution of computing has offered an opportunity to develop an understanding and shape our
environment. In last 1-2 years we have seen that the tipping point has reached where technology is
scalable to rapidly leverage the advancements in data science. Artificial Intelligence and Machine
Learning is disrupting the way the companies do business by reimagining the efficiency, speed and
functionality. In Supply Chain, Demand Forecasting and Planning has been the flag bearer in terms of
volume of data it manages. And intelligent automation can help the demand planners to boost their
Forecast Value Add process with good results. This project aims to demonstrate the value in using
Machine Learning techniques in Demand Planning process.

Keywords: Demand Planning, Forecasting, Machine Learning

13.3 Predict Foreclosure of Loan

The project involves predicting propensity of foreclosure based on lead indicators such as demographics,
transactional behaviour and performance on all credit lines. Predicting the probable timeline of the
foreclosure has also been included. Customer segmentation has further provided insights on the financial
and demographic splits between different sets of customers thus providing more insights. Prediction is
conducted by using machine learning techniques like Logistic Regression, Random Forest, Linear
Regression and k-means clustering. Significant factors causing Foreclosure of a loan account has been
identified and analysed using cost benefit analysis.

Keywords: Foreclosure, Home Loan, Balance Tenure, Customer Segmentation, Logistic Regression, Linear
Regression, Machine Learning

14.1 Case Outcome Prediction for SEBI Adjudication Orders using AI-ML Techniques

Predicting case outcome using machine learning models is one of the challenging problems in legal
domain. Lack of structured data, usage of handwritten texts, slow in adaptability of latest technology are
few of the challenges faced by data scientists working on this domain. Given technological advancement,
SEBI (Security Exchange Board of India) had started to maintain their cases online effective 2004. Our aim
is to build a Legal analytics solution that can assist SEBI and companies to find their chances of winning/
losing the case at higher accuracy. We use Natural Language Processing tools along with Machine
Learning techniques to build the solution. The proposed solution gives 95%+ test accuracy with very good
precision, recall and F1score.

Keywords: Case Outcome, Natural language processing, Machine learning techniques, Deep Learning,
Classification

14.2 Connecting Students and Industry

The primary focus of this project is to help TIEKE understand and categorise students based on their skill
and performance, also by analysing the current job market requirements in online platform we bring out
the key skill requirements that recruiters are looking for a fresher job profile. Therefore, by
understanding primarily both aspects the student capabilities and job market requirement, it would
enable TIEKE to bridge the skill requirement by providing right content and training for students. As a part
of their solution currently TIEKE is providing career counselling to students in colleges by analysing their
current academic performance with career choice and recommend industry orientation and skill gap
development programmes, the online job portal – FirstNaukri, Internshala and Freshersworld were web-
scrapped and the data was used to bring out insights in current skill requirements for fresher’s job
profile. Hence the project objective is to evaluate the student current academic performance along with
career interest and recommend them with skill enhancement content to fit into the job market
requirement that is currently trending.

Keywords: Job Market, Skill Requirement, Academic Performance, Industry Orientation

14.3 Building a Recommendation System for Tea Trails Nungambakkam

This project is aimed to use the business insights obtained at a Tea chain in Chennai, which despite having
good reviews has failed to bring in repeat customers according to the owner of the outlet. The franchisee
model has been successful in the north and western parts of India but is failing to generate more than 25
bills on daily basis. The study aims at identify the reason for this low sales volume and also attempts to
find out if there is pattern in the products consumed by the customers. This article applies a version of
market basket analysis (MBA) /association rule mapping principles to explore menu items assortments,
which are defined as the sets of most frequently ordered menu item pairs (Ping-Ho Ting, 2010).
The data collected has 3702 unique transactions covering approx. 9474 items for roughly 6 months. This
dataset suggested that guests ordering a given entrée would most likely then choose a particular side
dish to go with it. Findings point to significant opportunities for the client to use existing stored data to
better understand purchasing decision patterns that can significantly increase revenue per transaction.
Challenges to adoption and future research suggestions are offered. (Dolnicarc, 2016)
Along with finding out patterns the study also looks at rationalizing the menu options as the restaurant
has close to 250 items in its menu and it would not be prudent to save inventory for all of the items. The
study aims at using RFM methodology and eliminating the inventory which have not been sold for a long
time hence not adding up to the revenue of the organization. Indeed, Lee and MacGregor (1985) have
presented an analysis of search time that suggests that the optimal number of options per menu panel
over a wide range of conditions is in the range of only four to eight pages. (Kenneth R. Paap, 1986).
In this study we have found close to 100 items yield a low RFM score hence we have advised the client to
get rid of them. The report looks to demonstrate that it is possible to perform automatic sentiment
classification in the very noisy domain of customer feedback data. (gamon, 2004).
The study is made on the online reviews available on Zomato, Google and other online mediums in order
to identify customer perception towards the brand. We also carried out an SMS Blast to 24890
respondents yielding 360 responses through which we carried out a survey to cross reference the
attribute perception of the clients towards the store. Sentiment analysis/Text Analysis of online
restaurant reviews has confirmed the proposed underlying structure of online restaurant reviews (Qiwei
Gan, 2016).

Keywords: Franchisee Model, Market Basket Analysis, Decision Patterns, Sentiment Classification

15.1 Inventory and Average Closure Time Prediction for Consumer Cases in India

The general awareness among people on consumer rights has increased considerably in last few years.
Government also introduced necessary Acts and provisions to ensure that consumers can exercise their
rights in case of any fraud or deceit. All consumer cases across India, for the ease of access and
transparency, are available in CONFONET (Computerization and Computer Networking of Consumer Fora
in the Country) through NIC (www.confonet.nic.in). As per the data available on this forum, we can see as
people become more vigilant, an increase in number of consumer cases can be seen in past few years.
However, we think that this increase in cases has an impact both on lawyers as well as on consumers as
explained in the below statements which are also the problem statements for our study:
1. This increase in cases could lead to an impact on lawyers/law firms with regards to allocation of
lawyers. Hence the objective of our study with respect to this problem is to understand if we can identify
any pattern on the number of cases pending for judgement on month by month or quarter by quarter
basis. And if we can have a model to forecast the number of cases pending the judgement for a
month/quarter, lawyers/law firms can plan their resource allocation accordingly.
2. For consumers there is no way to identify the approx. closure time for their case. Hence, the objective
of the study with regards to this problem is to understand if we can have a model which can predict the
closure time for a consumer case based on the metrics like state, district and sector.

Keywords: Consumer Cases, Judgement, Closure Time Prediction

15.2 Patent Analysis based on innovations effectiveness in Pharmaceutical Industry

The objectives of this project are to devise a model that will help in deriving trends and insights
pertaining to the focus industry (Pharma industry), to forecast its patent granted effectiveness and
probable future trends of business losses due to rejection of patents and exhaustive study of the industry
fluctuations to understand how innovation helped the company become a top player.

Keywords: NLP, Patents grants and applications, R&D, Acquisitions, ROCE, Prediction, testing of model
accuracy

15.3 Personalised Promotions based on Customer Purchase Behaviour

A Grocery store has collected transactional data for households before and after rolling out a promotion
campaign. They want a model that is more customer-centric with a focus on customer success first and
eventually realizing business success as a result of this model. The data used is a public Dunnhumby’s
dataset, “The Complete Journey. This dataset tracks household spending over two years from a group of
2,500 households who have been identified as frequent shoppers. For anonymity purposes, the grocery
store chain has not been named. Tools used to achieve the objective of the project are R and Python to
apply datamining, predictive modelling and recommender techniques. Statistical tests, Market Basket
Analysis, RFM, cluster analysis and Recommender system were used. Our objective is to implement a
product bundling model, Segment customer base into various customer groups and build a
recommendation model from customer purchase behaviour to implement personalized promotions.

Keywords: Transactional Data, Customer-Centric, Predictive Modelling, Market Basket Analysis

16.1 Product Affinity Modelling and Sales Pipeline Enhancement Strategies

The purpose of the project is to segment the customers of a leading B2B IT product company through
RFM (Recency, Frequency and Monetary) modelling and to do a market basket analysis of its offerings so
that the sales teams can increase the sales through cross-selling and up-selling through these models.

Keywords: Customer Segmentation, Modelling, Market Basket Analysis

16.2 Decoding the Winner Formula on 22 Yards for Winner Prediction

Cricket is one the most watched sport now-a-days. Winning in cricket depends on various factors like
home ground advantage, performances in the past matches, experience of the players, performance at
the specific venue, performance against the specific team and the current form of the team and the
player. A lot of research has been done which measures the team ‘s performance and predicts the
winning percentage. This report briefs about the factors that cricket game depends on and discusses the
various methods which are used to predict the winning of a team in ICC World Cup. The study embarks
on predicting the outcome of ICC Cricket World Cup using a supervised learning approach. Study shows
team strength between the competing teams forms a distinctive feature for predicting the winner. The
data consist of team performances between each World cup with 160 variables with details of batting,
bowling, fielding performances, at home and away locations. Modelling the team strength boils down to
identifying and classifying team’s batting, fielding and bowling performances, forming the basics
approach. Statistics and recent performance of a team have been used to model Machine learning
algorithms used in predicting the outcome of the tournament.

Keywords: Cricket, Performance, ICC World Cup, Supervised Learning, Modelling, Machine Learning
Algorithms

16.3 SCM Analysis-Planning of E-Commerce Fulfilment Centre

With the advancement of technology, traditional business is moving in to online business called E-
commerce. E-commerce industry is growing at very fast rate. Consumers are attracted more towards E
commerce shopping due to various reasons such as high range of products to choose from, transparency
in prices, discounts and cash back offers, free product deliveries, after Sales services and return policy. E-
Commerce Industry has many challenges to overcome in order to grow and maintain the business. One of
the key challenges is maintaining high service levels to customers, as customers expect deliveries as early
as possible.
R was used for programming and churning out the data. Analysis is based on the confidential data from the
client. Data mining was done using R. Tableau was used to draw some graphs. Among analytical approach
& tools used, KNIME was also used for generating model outputs. K-Means & Hierarchical Clustering is
employed on the data to have a cluster dendogram. Microsoft Excel and Word is used to perform
calculation and collating the conclusions obtained during the project tenure. To draw the inferences from
the data, data analytics life cycle was followed starting from data collection, data cleaning, data exploration
and data analysis and finally drawing conclusion. Finally, this report also recommends the future works and
use cases that can be done on supply chain data.

Keywords: Data Churning, Data Mining, Model Outputs, K-Means, Hierarchical Clustering, Supply Chain
Data

16.4 Predicting Paid Conversions for TradeIndia.com

Online businesses are increasingly bearing high costs (Customer Acquisition Costs) and witnessing low
yields on customer acquisition and retention. This project is aimed at identifying key features (variables)
for one such web-based entity and attempts to construct a robust-positive-prediction model that can be
implemented to improve revenue realization and reduce costs - thereby impacting both the top and
bottom line.
At the same time, the focus throughout was to identify key (important) variables and list their importance
measures so the business management can build easy-to-implement, daily use rubrics for their
operations and sales functions.

Keywords: Customer Acquisition Costs, Retention, Prediction Model

17.1 Application of Machine Learning Techniques for Nifty Index Prediction

Stock market prediction has attracted much attention from academia as well as business. Due to the non-
linear, volatile and complex nature of the market, it is quite difficult to predict. As the stock markets grow
bigger, more investors pay attention to develop a systematic approach to predict the stock market. Since
the stock market is very sensitive to the external information, the performance of previous prediction
systems is limited by merely considering the traditional stock data. New forms of collective intelligence
have emerged with the rise of the Internet (e.g. Google Trends, Wikipedia, etc.). The changes on these
platforms will significantly affect the stock market. In addition, both the financial news sentiment and
Global Indices are believed to have impact on the Indian Stock Index NIFTY. In this study, disparate data
sources are used to generate a prediction model along with a comparison of different machine learning
methods. Besides historical data directly from the stock market, numbers of external data sources are
also considered as inputs to the model.
The goal of this study is to develop and evaluate a decision making system that could be used to predict
stocks short term movement, trend, and price. We took advantage of the public economic database
which allow us to explore the hidden information among these platforms. The prediction models are
compared and evaluated using machine learning techniques, such as Random forest, QDA, Logistic
regression and boosted tree. Numbers of case studies are performed to evaluate the performance of the
prediction system. From the case studies, several results were obtained: (1) the use of Global Indices
historical data sources along to improve the prediction performance; (2) the prediction models benefit
from the feature selection and dimensional reduction techniques. (3) The prediction performance
dominates the related works. Finally, a decision support system is provided to assist investors in making
trading decision on NIFTY.

Keywords: Predictive modelling, Data mining, Machining learning, Stock market, Time series, R, Tableau,
Capital market

17.2 Predictive Analysis for Abnormality in Wind Turbine

The objective of this project is to build a predictive model which can predict and also identify the trend in
values that can lead to an abnormality in wind turbine. There are various kinds of wind turbine
abnormality, generator bearing failure, gearbox failure, main bearing failure etc. Generator bearing
failure is a major issue and one measure which is in direct correlation to this failure is generator bearing
temp. In this project we are trying a build model to predict generator bearing temp as a collective
measure of the significant variables and try to identify the varying trend in the values which may serve as
a symptom of an upcoming abnormality. The IMR control charts are used to track the trend. Abnormality
can be defined as situations in which the turbine is no longer operating in the way it is intended for
maximum optimal power generation. To do this predictive analysis the pedagogy starts right from the
initial data gathered to the rightful identification of significant variables that are major influencers on the
running status of the wind turbine. The model identified can be used for other turbines in the same wind
farm and thus the whole farm is capable of predicting the abnormality.

Keywords: Predictive Model, Significant Variables, Maximum Optimal Power

17.3 Predicting Sales and Improving Profitability of Retail Chain

The project deals with analysis of retail transaction data of a retail chain located at four states of the
United States of America, that includes the sales, promotion information for multiple products and
brands with different categories for the past 156 weeks. The key challenge is to derive decision that
would enable the retail chain to identify the customer preferences out from the high volume of data.
The objective of the study to predict the sales for the future, address the relation between price and
sales. Also, estimate the impact on sales by changing the price gaps between items, promotions, displays
and feature. Also, address price cushion on increase in margin to retailer that would increase the sales. At
the end, considerable effort should be spent to identify the products that would enhance the sales and
margin to the retailer.
Keywords: Sales, Promotion, Products, Customer Preferences, Margin

17.4 Consumer Behaviour on Milk Consumption- A Comparative Study

This project is about Milk brands used across the states of Tamil Nadu. Through segmentation process we
segregate the consumers of the Milk Brands based on demographics factors and Behavioural factors.
Data capture from the consumers across the state is done through the Co-operative societies. The survey
data is collected randomly, not through any targeted forum or to specific set of people. Insights from this
project would help the Milk brands to understand their consumer expectation and target their consumer
specifically to improve ROI much better. India has always been the largest producer (an estimated 400
million litre per day currently) and consumer of milk in the world. But it remained a boring market largely
because the per capita consumption was low, and most of the milk was consumed in its basic, liquid
form, or at best as ghee and some butter. Over the past few years, though, a couple of things have
changed to make the market vastly more attractive to new players. India is termed as an ‘Oyster’ in the
world dairy business as its revenue growth is exponential, as of 2017 revenue stands in 80,000 crores.
With population segment like India, where most of its nutrients depends on Milk & Milk based product,
market is wide open for lot of investment from private brands because most of the Milk production in
India is from the village co-operative societies.

Keywords: Segmentation, Demographics, ROI, Revenue, Investment