Scribd
Upload
110 views

ORM-2 Assignment 1

The document summarizes the recommendations provided to iD fresh foods to help solve their business problems related to forecasting demand and reducing returns through better analytics. It analyzes the available sales data, identifies missing values and outliers, recommends techniques for feature engineering and model building. Based on the analysis, exponential smoothing is identified as the best forecasting model for a specific beat and SKU. Final recommendations include training sales teams to collect better data, implementing the exponential smoothing model, and continuing to test and improve models over time with more comprehensive data.

Uploaded by

Komal Modi
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as DOCX, PDF, TXT or read online on Scribd
110 views

ORM-2 Assignment 1

The document summarizes the recommendations provided to iD fresh foods to help solve their business problems related to forecasting demand and reducing returns through better analytics. It analyzes the available sales data, identifies missing values and outliers, recommends techniques for feature engineering and model building. Based on the analysis, exponential smoothing is identified as the best forecasting model for a specific beat and SKU. Final recommendations include training sales teams to collect better data, implementing the exponential smoothing model, and continuing to test and improve models over time with more comprehensive data.

Uploaded by

Komal Modi
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as DOCX, PDF, TXT or read online on Scribd
You are on page 1/ 2

Komal Modi | BJ21146 | Section C

ORM-2 Assignment 1
1. What is the business problem that ‘iD fresh foods is trying to solve and what are the
corresponding analytics problem?
ID fresh foods was getting affected because they had not applied analytical tools. They faced
the business problem of forecasting the accurate quantity of products to be delivered to each
retail outlet while ensuring that returns due to expiry date are the least. The company also did
not want to lose business in case their products weren’t available in stores. Correspondingly ID
fresh foods relied on the salespeople’s intuition to predict sales and they did not keep
information about returns and stockout due to operational inefficiencies. This shows the
company did not have a robust analytical model in place to predict the sales and they couldn’t
service the retail outlets effectively.
2. The dataset has missing information. Given the context of the problem, what is the best
strategy to handle missing information?
The dataset given to us does not having information till the end of November. There are some
ways that can help. We could either remove the datapoint which has the missing value, but we
already have less amount of date. Another way is to fill the missing value with a global constant
provided we have a way of determining the global constant, but we don’t. And lastly, we can
replace the missing data point with measures of central tendency like mean, median and mode
depending on the distribution of the dataset. Since the current dataset is almost normally
distributed, we could go ahead and replace the missing values with the mean.
3. Check whether there are outliers in the data. How do we handle outliers in the data of this
case?
There are outliers in beat 98. For our dataset outliers can be handled using quantile-based
flooring and caping. This can be done since the outliers are not very distant from the lower and
upper bounds.
4. Discuss the approach or strategy for variable selection and feature engineering that may be
used to build analytical model in this case.
Forward selection or backward elimination can be used to choose variables that influence
actual demand for iD items. In forward selection, variables are added one by one to reduce
error, whereas in backward elimination, all elements are assessed at the same time and then
removed one by one until the allowable error is met. Weather conditions, days of the week,
festival periods, holidays, and outlet location are all factors that can influence actual demand
for iD items.

5. Forecasting model can be developed at individual store level or at beat level. What will be
your suggestion to ‘iD fresh about this?
Komal Modi | BJ21146 | Section C

Since the stores are located along comparable routes, the beat level technique would be better
because:
 Stores are exposed to bigger volatility as they may be subject to client shuffling.
 As previously stated, some establishments are part of a large chain of stores or a
superstore and are therefore more susceptible to price swings.
 Maintaining precise beat-level data is easier. A salesperson may be unable to
appropriately capture store-specific data.
 Because the distribution would be at the beat level, forecasts made at the beat level
would be more useful in filling the vehicle with the needed quantity.

6. Develop forecasting models using techniques for Beat 81 and SKU 95.  Select the most
appropriate technique using mean absolute percentage error (MAPE)
The results of all the model built are as follows -

Method Naive 2DMV 3DMV 4DMV Exponenti Regressio


al n

MAPE of 31.46 28.02 24.92 23.83 20.73% 20.16%


train % % % %
MAPE of 32.04 29.87 26.13 25.25 22.52% 23.67%
test % % % %

It is seen that the regression model is the best fit for anticipating demand for Beat 81 and SKU
95 based on MAPE. However, while testing models using test data, we discover that the
Exponential smoothening model outperforms the Regression model.
7. What should be the deployment strategy for ‘iD fresh food?
Firstly, the sales team should be trained to collect relevant information from the stores in their
beat. It should include data like product demand, reason and return of product, frequency and
quantity of return etc. This data should be input into a database regularly. This will make the
analytical model more accurate. Feedback about returns and demands should be also taken
from the salespeople frequently.
8. What will be your final recommendations to ‘iD?
The exponential smoothening model, which had the lowest MAPE of all the models, is the final
recommendation to ID. However, it should aim to collect more data and integrate more
variables to better estimate demand, and it should continue to test which model outperforms
the others in the long run.

You might also like