E X P L O R AT O RY D ATA A N A LY S I S

IN BFS

Presented by:
Shakti Singh
Introduction to Exploratory Data Analysis
• “While fraud reduction is a common goal for banks and financial institutions, analytics
can be used to manage risk instead of simply detecting fraud” (Janaha, Data Analytics
in banking and Financial Services 2023).
• “Analytics can be used to identify and rate individual customers who are at risk of
fraud and then apply different levels of monitoring and verification to those accounts.
Analyzing the risk of the accounts allows banks and financial institutions to know
what to prioritize in their fraud detection efforts” (Janaha, Data Analytics in banking
and Financial Services 2023).
• “With the rise of computing power and new analytical techniques, banks can now
extract deeper and more valuable insights from their ever-growing mountains of data.”
Moreover, “The recent dramatic increases in computing power have allowed banks to
deploy advanced analytical techniques at an industrial scale” (Dash et al., Risk
analytics enters its prime 2017).
Introduction to Exploratory Data Analysis

• “Machine-learning techniques, such as deep learning, random forest, and XGBoost, are now
common at top risk-analytics departments” (Dash et al., Risk analytics enters its prime 2017).
• “The new tools radically improve banks’ decision models. And techniques such as natural-
language processing and geospatial analysis expand the database from which banks can
derive insights” (Dash et al., Risk analytics enters its prime 2017).
• “This means that risk teams can increasingly measure and mitigate risk more accurately and
faster” (Dash et al., Risk analytics enters its prime 2017).
• “Banks that are fully exploiting these shifts are experiencing a “golden age” of risk analytics,
capturing benefits in the accuracy and reach of their credit-risk models and in entirely new
business models” (Dash et al., Risk analytics enters its prime 2017).
Introduction to the Case Study

• “The loan providing companies find it hard to give loans to the people due to their insufficient or
non-existent credit history. Because of that, some consumers use it to their advantage by becoming a
defaulter” (upGrad, Credit EDA Assignment 2023).
• “When the company receives a loan application, the company has to decide for loan approval based
on the applicant’s profile” (upGrad, Credit EDA Assignment 2023).
• “This case study aims to identify patterns which indicate if a client has difficulty paying their
instalments which may be used for taking actions such as denying the loan, reducing the amount of
loan, lending (to risky applicants) at a higher interest rate, etc.” (upGrad, Credit EDA Assignment
2023).
• Problem Statement: “The company wants to understand the driving factors (or driver variables)
behind loan default, i.e. the variables which are strong indicators of default. The company can
utilise this knowledge for its portfolio and risk assessment” (upGrad, Credit EDA Assignment 2023).
Procedure followed from the Beginning
• All the warning and necessary libraries were imported such as import warnings, numpy,
pandas, matplotlib.pyplot and seaborn.
• Then the CSV file called ‘application_data.csv’ was imported along with CSV file
‘previous_application.csv’.
• ‘application_data.csv’ was converted to a pandas data frame and stored in a variable and same
was done for ‘previous_application.csv’.
• First analysis was done on current application or application data set and then later it was
done on the previous application dataset.
• File was read using the head( ) and tail( ) function
• After that the columns were checked, and info was printed.
• Then the data set was checked for null values.
Null Value Treatment
• Then the null values were converted to a percentage format.
• Columns that were having null values more than 35 percent were removed from the dataset.
• Then null values that were less than 13 to 20 percent that were missing in the dataset were
identified labeled as minor_missing_values.
• Then the unique values of the columns were identified using the nuinque ( ) function and was
determined that they were of categorical nature.
• Some unnecessary columns were dropped that were not needed for the analysis.
• Then data type correction was performed making columns that are of object type or
categorical or numerical type and were converted accordingly.
• Columns having negative values were converted to positive where it was required.
Data Binning

Inference: The graphs shows that range of the amount of money

borrowed by males and females. In this case females are the major
borrowers. The most borrowed amount of money by both male and
female falls in the range of 250000 to 500000.
Outlier Identification and Suggested Treatment
Box Plot for CNT_CHILDREN

Inference: We can
clearly see that
there are many
outliers present, we
can use median or
mode to impute the
outliers because
mode or median
will give an
accurate
representation of
the whole dataset.
Box Plot for AMT_INCOME_TOTAL
Inference: We can
observe that there are
outliers present here
as well. We can use
median or mode to
impute the outliers
because mode or
median will give an
accurate
representation of the
whole dataset
Box Plot for AMT_CREDIT

Inference: There are outliers

present as well. We can use
median or mode to impute the
outliers because mode or
median will give an accurate
representation of the whole
dataset
Box Plot for YEARS_EMPLOYED
Inference: We can see
that there is an outlier
here as well which is at
1000 years realistically it
is not possible. we can
use median or mode to
impute the outliers
because mode or median
will give an accurate
representation of the
whole dataset
Box Plot for AMT_REQUIRED_CREDIT_BUREAU_QRT
Inference: There is an
outlier here as well. We can
use median or mode to
impute the outliers because
mode or median will give an
accurate representation of the
whole dataset
Heat Map of Target Variable Zero
Inference: We can see clearly that
our upper matrix is symmetrical
and the amount credit, and
amount goods price have the
highest positive correlation 0.99
which means they are positively
linearly correlated. However, the
ext. source 3 and amount total
income have a negative
correlation. Additionally, the
amount good price and amount
annuity are also presenting with a
good correlation.
Heat Map of Target Variable Zero
Inference: We can see from the
above heat map that the amount
credit and amount goods price
have highest positive correlation
which shows the positive linear
relation of the two variables.
However, ext. source 3 and
amount income total have a
negative correlation. This just
shows that two variables that
have negative correlation are not
necessarily dependent on each
other.
Numerical Univariate Analysis
Box Plot comparing
YEARS_BIRTH by using clients
from dataset of Target 0 and 1

Inference: From the box plot we

observe that clients without
payment difficulties are of the
age 35 to 55 years , And clients
with payment difficulties are
between ages of 32 to 48 years.
Also, the clients that do not have
difficulty paying have a higher
IQR.
Numerical Univariate Analysis
Box Plot comparing
YEARS_EMPLOYED by using
clients from dataset of Target 0
and 1

Inference: The box plot shows

that clients without payment
difficulties have been employed
between 2 to 9 years whereas,
clients with payment difficulties
have been employed between 2
to 7 years.. Also, the clients that
do not have difficulty paying
have a higher IQR.
 Box Plot comparing
Numerical Univariate Analysis AMT_GOODS_PRICE: clients
from dataset of Target 0 and 1
We can see that clients who have
not had difficulty paying the
loans have been granted a lower
price when compared to the
clients that have had difficulties
in making the payments. To put it
numbers, clients that didn't have
difficulties they have been
granted the loan at a low cost
of .25 to 1.25 whereas for clients
that have had difficulties they
have been granted also similar
rates of interest. Now this can be
a result of data not being clean as
expected but generally people
with payment difficulties are not
granted loans at such low
percentage and if they are
granted, they are usually higher.
Numerical Univariate Analysis
Box Plot comparing
YEARS_ID_PLUBLISHED and
clients from dataset of Target 0
and 1

It is an interesting observation
here one can see that clients
that have been paying
regularly they have moved
more frequently and clients
that have had problems with
payment have moved less
often.
Numerical Univariate Analysis
Box Plot comparing
AMT_ANNUITY and clients
from dataset of Target 0 and 1

We can clearly observe that

both clients with and without
difficulties have almost the
same Annuity but there is a
difference in their IQR,
clients that have had
difficulties have a higher IQR
when compared to clients that
haven't had problems.
Categorical Univariate Analysis
Count Plot comparing
CONTRATC_TYPE and clients from
dataset of Target 0 and 1

Both the client base people

with difficulty or without
have taken more cash loans
rather than the revolving
loans. These depend on many
factors such as the income of
the populous that have the
greatest number of people in
that category or their
education.
Categorical Univariate Analysis
Count Plot comparing
CONTRATC_TYPE and clients
from dataset of Target 0 and 1

An interesting observation,
Males in both the categories
have defaulted less than the
females. However, the
number of females that have
taken a loan is also more than
males.
Categorical Univariate Analysis
Count Plot comparing
EDUCATION_TYPE and clients
from dataset of Target 0 and 1

Here we can see that most of

the population or the database
has a secondary/secondary
special education, and this
could be a driving factor that
we were seeing earlier with
people taking more cash loans
rather than revolving loans.
Categorical Univariate Analysis
Count Plot comparing
HOUSING_TYPE and clients
from dataset of Target 0 and 1

Both the categories of

customers have their own
houses. But we can also see
that people with difficulties
are way less than people
without difficulties, which is
good news. As less defaulters
are present, and bank can
make recovery of their money
by asking the defaulters to
keep their houses as
collateral.
Categorical Univariate Analysis

Count Plot comparing

OCCUPATION_TYPE and
clients from dataset of Target 0
and 1
Categorical Univariate Analysis
We can see in the count plot OCCUPATION_TYPE in previous slide that in both categories that is people
who have difficulty paying the loans and people who don't have difficulty paying the labor class is at the
highest because they are mostly dependent on daily wages and seasonal work and when they don't get paid,
they default However, when they do get paid, they tend to pay back because of the small loan amount has
small installment as well. Additionally, we can see that the sales category has the second highest numbers
because they are there in majority as in the number of people who are in the sales field are comparatively
more in numbers. Lastly, we can see that employees from the IT sector and HR sector are the least people to
default as they have better paying jobs and less need for borrowing money.
Bi-variate Analysis
Scatter Plot on Years
Employed and Total
Income

In this plot we can

observe that people
with payment
problems are more
consistently in the
low-income group
that may be the
result of the labor
and sales category.
However, people
with less payment
difficulty are ranging
from low income to
high income.
Bi-variate Analysis
Scatter Plot on Credit
amount and Goods
Price

Here we see the

positive correlation
of amt credit to
goods price in both
the cases.
Bi-variate Analysis
Scatter Plot on Credit
amount and Annuity

Inference:
People without
payment difficulties
take more credit for
the annuity
Bi-variate Analysis
Count Plot on Contract Type
and Credit Range

Inference:
Here we see that
people from both the
category take loans
of cash type more
than the revolving
loans but that can be
credited to most
people in Labor and
sales class.
Bi-variate Analysis
Count Plot on Gender and
Credit Range

Inference:
In both categories
females have taken
out more loans and
even the amount is
greater in both cases
in females.
Numerical and Categorical Bi-variate Analysis
Box Plot on Credit
Amount and Education
Type

Inference: In case of
clients with difficulty
paying the loan, we can
see that people with
higher education
struggled more in
repayment it can be
because of the situation
of job market and the
amount of loan or their
employment status. On
the other hand, clients
who don't have
payment difficulty is
also leading with higher
education.
Numerical and Categorical Bi-variate Analysis
Box Plot on Total
Income and Education
Type

Inference:
In both cases there are
numerous outliers, but
academic degree is the
least in both the cases
because of the client
base of the dataset
which is less.
Numerical and Categorical Bi-variate Analysis
Box Plot on Credit
Amount and Occupation
Type

Inference:
We can observe that the
amount of credit taken
is more in clients with
no difficulties and
defaulters tend to take
less amount of credit in
any occupation type.
Also, we can see that
Accountants and
Mangers tend to take
more loans and have
more difficulty paying
back as well.
 Previous Application Data

• We have followed the same steps as the current application data for Data Cleaning.

• After Cleaning the data, we have merged both the current and previous application data to perform the final
analysis.
Final Dataset Analysis
Count Plot and Pie Plot showing Different status of Loan Offered
Final Dataset Analysis
Count Plot for Contract Type with four subcategory
Inference:

We can see that the

Cash loans are more
commonly taken
than the revolving
loan. This can be
since Laborer and
low-income type
people are in
majority in the
dataset.
Final Dataset Analysis
Count Plot for Client Type with four subcategory
Inference:

In all the cases above

the repeating
borrower is getting
rejected more often
followed by new
applicants
Final Dataset Analysis
Count Plot for Gender with four subcategory
Inference:

In all the cases

females have been
refused more as we
have seen above it is
because they are
more in numbers
when compared to
males hence the
disparity
Final Dataset Analysis
Count Plot for Education Type with four subcategory
Inference:

Here people from

less educated
background that is
secondary and
secondary special
have taken more
loans compared to
others
Final Dataset Analysis
Count Plot for Income Type with four subcategory
Inference:

In all cases working

people have been
approved more and
have more loans
when compared to
other classes of
people followed by
commercial
associates
Final Dataset Analysis
Count Plot for Family Status with four subcategory
Inference:

We can observe that

married clients are
the biggest
borrowers from the
bank.
Final Dataset Analysis
Count Plot for Payment Type with four subcategory
Inference:

Cash through Bank

category is the
preferred form of
borrowing amongst
all the clients.
Final Dataset Analysis
Count Plot for Occupation Type with four subcategory

Inference:

Laborer, sales staff,

core staff and drivers
are the leading
borrowers who
highest in approval
and rejection.
Univariate Analysis on Final Dataset
Logarithmic Comparison of Purpose of Borrowing

Inference:

Majority of rejected
loans are from the
category 'repairs'.
Also, education has
equal number of
approves and
rejection
Paying other loans
and buying a new car
is having significant
higher rejection than
approvals.
Univariate Analysis on Final Dataset
Logarithmic Comparison of Contract Status

Inference:

As we can see that there is disparity between the

two categories target 0 and 1 but loan purposes of
'Repairs' are the highest amongst all purposes.
There are few places where loan payment is
significantly higher than facing difficulties.
They are 'Buying a garage', 'Business development',
'Buying land', 'Buying a new car' and 'Education'
Hence, we can focus on these purposes for which
the client is having for minimal payment
difficulties.
Bivariate Analysis on Final Dataset Inference:
Logarithmic Comparison of Credit Amount
The credit
amount of
Loan purposes
like Buying a
home, land and
car and
Building a
house is higher.
State servants
have a
significant
amount of
credit applied.
Money for
third person or
a Hobby is
having less
credits applied
for.
Bivariate Analysis on Final Dataset Inference:
Logarithmic Comparison of Credit Amount vs Housing type
The Housing type, office
apartment is having
higher credit for target 0
and co-op apartment is
having higher credit for
target 1. Thus, it is safe
to conclude that bank
should avoid giving
loans to the housing type
of co-op apartment as
they are having
difficulties in payment.
Bank can focus mostly
on housing type with
parents or House or
apartment or municipal
apartment for successful
payments.
Conclusion on Final Dataset
1. Bank should focus on making cash loans more accessible as majority of the people are opting for that also the revolving loans have equal
number of approval and refusal which makes the revolving loan less desirable.
2. Banks should also focus on new clients as they are more likely to get approved and less likely to default.
3. Banks should focus more on female candidates as they are greater in number when it comes to applicants.
4.People from secondary background should be the target area where banks should concentrate as they are the leading majority followed by
higher education people.
5.Working class people and commercial associates are the highest number of people who should be targeted but they should also put more focus
on Pensioners and state servants as they are the new emerging majority who are approved more often and are less likely to default.
6.Married people are leading in the majority when it comes to approval and rejection, but banks should focus on single people separated and
widowed.
7. As stated earlier, cash through bank is a leading form or loan giving but bank should also focus on non-cash loan from bank accounts.
8. When we talk about the occupation type the banks should keep in mind the IT and HR staff as they are the least likely to default and target them
the most.
9.The banks should shift their focus from urgent needs as they have no unused offers that means they are more likely to have more defaulters.
Conclusion on Final Dataset
10. Banks should avoid giving loans to the housing type of co-op apartment as they are having difficulties in payment. Also, Bank can focus
mostly on housing type with parents or House/apartment or municipal apartment for successful payments.
11. Banks can also focus on loan purpose of buying house or apartments, cars and additionally, they can come out with creative schemes that helps
loan applied for third person as that purpose has significantly low amount of credit applied.
References
• upGrad. (n.d.). Credit EDA Assignment. https://learn.upgrad.com/
• Dash, R., Kremer, A., Nario, L., & Waldron, D. (2017, June 6). Risk analytics enters its
prime. McKinsey & Company.
https://www.mckinsey.com/capabilities/risk-and-resilience/our-insights/risk-analytics-enters-
its-prime
• Janaha, V. (2023, May 3). Data Analytics in banking and Financial Services. Zuci Systems.
https://www.zucisystems.com/blog/how-is-data-analytics-used-in-finance-and-banking-
sector/#:~:text=While%20fraud%20reduction%20is%20a%20common%20goal%20for,levels
%20of%20monitoring%20and%20verification%20to%20those%20accounts.
THANK YOU