CREDIT EDA CASE STUDY

PGDDS C26 November 2020

Presented By
VADAGAM PRAVALIKA
 Introduction

1 Objective

• When the company receives a loan application, the company must decide for loan approval based on the applicant’s profile. Two types of risks are
associated with the bank’s decision:

• If the applicant is likely to repay the loan, then not approving the loan results in a loss of business to the company

• If the applicant is not likely to repay the loan, i.e., he/she is likely to default, then approving the loan may lead to a financial loss for the company.

2 Data set used

• Current applications “application_data.csv”

• Previous applications “previous_application.csv”

 Steps Involved

1 Loading Data 8 Bivariate Analysis : Numerical – Numerical

2 Inspecting Data 9 Correlations

3 Data Analysis 10 Loading Data "previous_application.csv"

4 Checking the Data Imbalance 11 Final Conclusions

Univariate analysis for Categorical columns

5

6 Univariate Analysis for Numerical columns

7 Bivariate Analysis : Numerical – Categorical

2.Loading Data / EDA
 EDA ANALYSIS

• Reading Data

1. Reading Dataset application_data.csv.

• We found that the column CODE_GENDER having value ‘XNA’ which simply
2. Reading Dataset previous_application.csv.
means Not Available . So, we imputed those values with the most frequent
value(mode value) i.e., F.

• Inspecting Data frame

• Then we found that column ORGANIZATION_TYPE also have 18% ‘XNA’
1. Inspecting and understanding Data values. So firstly, we checked that whether the values are Missing Completely at
Random(MCAR), Missing at Random(MAR) or Missing Not at
• Checking few records of Dataset such as .shape, .info(), .describe() . Random(MNAR).After comparing the values of ORGANIZATION_TYPE with third
2. Data Cleaning variables i.e., values of column NAME_INCOME_TYPE , we found that Clients
• Checking the percentage of null values in the data frame in descending order. who are pensioner were having XNA values in ORGANIZATION_TYPE. So, we
• Analyzing number of null columns replaced “XNA” with “Pensioner”.
• Dropped columns having null values > 35%
• Imputing Columns having null values ≤ 19% with Mode values for numeric columns except for
continuous numeric columns we imputed with Median value. • Similarly, we found the same relation between OCCUPATION_TYPE and
NAME_INCOME_TYPE , so we imputed null values with “Pensioner”.
3. Handling Errors in Data types and Data

• Checking the percentage of null values in the data frame in descending order.
• Analyzing number of null columns : 49
• For convenience to our analysis, we made binning based on the quantiles for
• Dropped columns having null values > 35%
the columns mentioned below:
• Imputing Columns having null values ≤ 19% with Mode values for numeric columns except for
AMT_INCOME_TOTAL , AMT_CREDIT AND AGE_DAYS
continuous numeric columns we imputed with Median value.

• After observing data frame, we find columns: 'DAYS_BIRTH', ‘DAYS_EMPLOYED',

‘DAYS_REGISTRATION', 'DAYS_ID_PUBLISH‘ and 'DAYS_LAST_PHONE_CHANGE’ which had
negative or mixed values, So we imputed them with absolute values for our analysis.
• Then we changed the values of columns FLAG_OWN_CAR and FLAG_OWN_REALTY from ‘Y’ and
‘N’ to 1 and 0 respectively for convenience in analysis.
3.Analysis
3.1 Data Types Conversion's for better understanding of Variables

In data frame, we observed some columns having d-type

“object” ,which can be converted to d-type “category”
which will be convenient for our analysis as well as it will
save memory usage.
There are so many columns in data frame. We’ll remove
columns which we don’t need for further analysis.
 3.2 Finding and Analyzing outliers

We made list of all numerical columns .And then

plot Boxplots for each numerical columns.

• IQR for AMT_INCOME_TOTAL is very slim, and it have many

outliers. It has Maximum value of 117000000 which is a huge
variation from 75th percentile.

AMT_INCOME_TOTAL AMT_GOODS_PRICE

• Third quartile of AMT_GOODS_PRICE, AMT_CREDIT is larger

as compared to First quartile which means that most of
the Credit amount of the loan of customers are present in the
third quartile. Here maximum value is 4050000 which varies a lot
from 75th percentile.

• DAYS_EMPLOYED have maximum value at 375000 ,which vary a

lot from mean and 75th percentile.
AMT_CREDIT DAYS_EMPLOYED
• Visual Representation by Boxplot is shown
4. Checking Imbalance
 4.1 Checking Data Imbalance for Target Variable

Since there is a huge imbalance between

the TARGET variables 0 and 1, it makes
more sense to divide data frame into
two sub datasets then continue our
analysis.

We have splits data frame as

follows:

• Target0 : (Non-Defaulted Population)

Clients without Payment Difficulties.

Conclusion
• Target1 : (Defaulted Population)
Clients with Payment Difficulties.
We have found that -
Ratio of Data Imbalance is “11.3”
In order to analyze the imbalance and various aspects of data we will perform
various types of analysis such as:

• Univariate analysis , Bivariate analysis , Multivariate analysis

5.Univariante Analysis
Categorical
 5.1 Gender/Age Group with respect to Target variables

• It seems like Female clients applied

higher than male clients for loan.

• 66.6% Female clients and 33.4%

male clients are payment difficulties.

• 57% Female clients and 42% male clients

are with payment difficulties.

• Middle Age Group(35 – 60) have

applied most and have higher payment
difficulties amongst all.

• While Senior Citizens(60 -

100) and Very young (19 - 25) age
group facing fewer paying difficulties as
compared to other age groups
 5.2 CONTRACT/INCOME Type with respect to Target variables

• Most of the clients applied for the Cash

Loan while a very small proportion
applied for Revolving Loan.

• But Clients applied for Cash Loan have

higher payment difficulties.

 Clients who applied for loans were getting income by Working, Commercial associate
and Pensioner are more likely to apply for the loan, highest being the Working-class
category .

 Businessman, students and Unemployed less likely to apply for loan .

 Working category have high risk to default.

 State Servant is at Minimal risk to default.

 5.3 Education/Family Type with respect to Target variables

• Clients having Secondary/Secondary Special Education are more

likely to apply for the Loan.

• Also, clients having Secondary/Secondary Special Education are

facing higher payment difficulties ,so they have high risk to default.
Other education types are at minimal risk

• It seems that Married clients applied most for the loan and have higher
payment difficulties.

• Widows are less likely to apply for the loan and have minimal risk.

• Clients with the single relationship have minimal risk to default i.e., have less
payment difficulties.
 5.3 Occupation/Income Range with respect to Target variables

• Pensioners and Laborers have applied the most for the

loan.

• Pensioner being highest followed by laborers have higher

payment difficulties , so have high risk to default.

• Clients having Medium salary range are more likely to apply for the loan.
And have higher payment difficulties.

• Clients having low and medium income are at high risk to default.

• Clients having high salaries are at minimal risk.

 5.3 Housing Type/Credit Range with respect to Target variables

• Most of the clients applied for the loan owns a

house/apartment and have a higher payment difficulties.

• Other clients have less payment difficulties with low

applications for loan.

• Most of the clients applied for Medium Credit Amount

for the loan .

• Clients applying for medium and low credit have high

payment difficulties and have high risk to default.
6.Univariante Analysis
Numerical
6.1 Annuity/Credit with respect to Target variables

• Distribution of
AMT_ANNUITY for
Target1 is broader than
Target0

• Distribution of AMT_CREDIT for

Target1 and Target0 is similar.
6.2 Goods Price/ Amt Income with respect to Target variables

•Distribution of
AMT_GOODS_PRICE for
Target1 and Target0 is
similar.

• Distribution of _INCOME_TOTAL for

Target1 and Target0 is similar.
 6.3 Points To Conclude

Notes

• People with target one has largely staggered income as compared to

target zero. Dist. plot clearly shows that the shape in Income total,
Annuity, Credit and Good Price are similar for Target 0 and similar for
Target 1.

• The plots are also highlightingthat people who have difficulty in paying
back loans with respectto their income, loan amount, price of goods
against which loan is procured and Annuity.

• Dist. plot highlights the curve shape which is wider for Target 1 in
comparison to Target 0 which is narrower with well defined edges.
 7. Bivariate Analysis
Numerical V/s Categorical
7.1 Income_Amount Vs Education_Status Vs FAMILY_Status among Target Variables

• Widow Client with Academic degree have a very few outliers and doesn't have First and
Third quartile. Also, Clients with all type of family status having academic degree have
very less outliers as compared to other type of education.

• Income of the clients with all type of family status having rest of the education type lie
Below the First quartile i.e., 25%.

• Income amount for Married clients with academic

degree is much lesser as compared to others.

• (Defaulter) Clients have relatively less income as

compared to Non-defaulters.
7.2 Credit Vs Education_Status Vs FAMILY_Status among Target Variables

• Clients with all Education type except Academic degree have large number of outliers

• Most of the population i.e., clients credit amounts lie below 25th percentile.

• Clients with Academic degree and who is a widow tend to take higher credit loan.

• Some of the clients with Higher Education, Incomplete Higher Education, Lower
Secondary Education and Secondary/Secondary Special Education are more likely to take
high amount of credit loan.

• Married client with academic applied for higher

credit loan. And doesn't have outliers. Single clients
with academic degree have a very slim boxplot with
no outliers.

• Some of the clients with Higher Education,

Incomplete Higher Education, Lower Secondary
Education and Secondary/Secondary Special
Education are more likely to take high amount of
credit loan.
 8. Bivariate Analysis
Categorical V/s Categorical
 • Though count of working clients applying for loan is significantly high , risk to default in payments is less
as compared to others

• Count of clients with income type Maternity leave is only 5, but risk to default in payments for those is
minimum among all the
income types.

• Same condition is observed in case of unemployed. Though count is very low, risk to default in payments
is low.

• Pensioner, State servant and Commercial associate have higher risk to default.

• Cash loans have higher risk to default, Revolving loans have comparatively lower
risk for the same .
 • Clients having Low
income have high risk
to default followed by
clients with medium
and very low income.

• Clients with high

salaries have minimal
risk to default.

• Clients having
Academic Degree and
higher Education have
lower risk to default.

• Clients having Lower

• Low-skill Laborers have higher risk to default
Secondary ,
Secondary/Secondary
Special Education have
very high risk to
• Managers, High skill tech staff and Accountants have
relatively lower risk to default
default.
8.Correlation’s
 Points to Mark:
GOODS PRICE Vs CREDIT AMOUNT
1. High correlated variables
for both defaulters and non
T-0 T-1 - defaulters. So as the home
price increases the loan
amount also increases

2. High correlated variables

for both defaulters and non
defaulters . So as the home
CREDIT AMOUNT Vs ANNUITY price increases the EMI
amount also increases
which is logical
T-0 T-1

Conclusion

• All three variables Are

highly correlated for both
ANNUITY Vs GOODS PRICE defaulters and non-
defaulters, which might not
give a good indicator for
defaulter detection
T-1
T-0
8.2 Correlations between numerical variables using "Heatmap's"

TARGET 0 As we can see from correlation heat map

for TARGET-0, There are number of
observation we can point out

• AMT_CREDIT is inversely proportional to the

DAYS_BIRTH, peoples belongs to low-age group
taking high Credit amount and vice-versa

• AMT_CREDIT is inversely proportional to the

CNT_CHILDREN, means Credit amount is higher
for less children count client have and vice-versa.

• AMT_INCOME_TOTAL is inversely proportional to

the CNT_CHILDREN, means more income for less
children client have and vice-versa.

• Less CNT_CHILDREN client have in densely

populated area.

• AMT_CREDIT is higher to densely populated area.

• AMT_INCOME_TOTAL is also higher in densely

populated area.
8.2 Correlations between numerical variables using "Heatmap's"

• AMT_CREDIT is inversely
TARGET 1 proportional to the DAYS_BIRTH,
peoples belongs to low-age group
taking high Credit amount and vice-
versa Conclusion
This heat map for Target 1 is
• AMT_CREDIT is inversely also having quite a same
observation just like Previous
proportional to the CNT_CHILDREN,
Target 0. But for few points
means Credit amount is higher for
are different. They are listed
less children count client have and
below.
vice-versa.

• AMT_INCOME_TOTAL is inversely • The client's

proportional to the CNT_CHILDREN, permanent address
means more income for less children does not match
client have and vice-versa. contact address are
having less children.
• Less CNT_CHILDREN client have in
• The client's
densely populated area.
permanent address
• AMT_CREDIT is higher to densely does not match work
populated area. address are having
less children.
• AMT_INCOME_TOTAL is also higher
in densely populated area.
9. Top 10 Correlation’s
 9.1Correlation Remarks

• Top 10 correlations between both Default (TARGET 1) and non default(TARGET 0)

clients are almost at the same level for different variables

TARGET 0 TARGET 1

V/s
10. Loan Distributions and
Purposes
 10.1 Percentage of `NAME_CONTRACT_STATUS` and `NAME_CLIENT_TYPE`

• Around 80.7% clients were repeaters

applying for loan.

• 14.5% clients are new applying for the

loan.

Percentage of contracts approved or not in previous

applications

• Approved :- 38.8%
• Refused :- 58.5%
• Canceled :- 2.3%
• Unused offer :- 0.31%
10.2 CONTRACT_STATUS V/s LOAN_PURPOSE

• Most rejection of loans came from

purpose "Repairs".

• For "Education" & "Medicine"

purposes we have equal number of
approves and rejection

• “Paying other loans” and "Buying a

new car'` is having significant higher
rejection than approves.
10.3 TARGET V/s LOAN_PURPOSE

• Here also there is a high

variation for “Repairs" for both
the targets

• Comparing "Education" &

"Medicine" purposes , Medicine
there is high no. of clients
having difficulties for re
payment the loan amount
compare to "Education“

• “Buying used Car" and “Building

purpose” client having
difficulties in payment have
equal ratio
10.4 AMT_CREDIT vs NAME_HOUSING_TYPE

Here for Housing type

• Office apartment is having higher credit of target

0 and co-op apartment is having higher credit of
TARGET 1.

Conclusion

So, we can 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
municipal apartment for successful payments.
Conclusions
 Final Insights

• We observe a decrease in the percentage of Payment Difficulties who are pensioners and an increase in the
percentage of Payment Difficulties who are working when compared the percentages of both Payment
Difficulties and non-Payment Difficulties.

• We observe a decrease in the percentage of married and widowed with Loan Payment Difficulties and an
increase in the the percentage of single and civil married with Loan Payment Difficulties when compared with
the percentages of both Loan Payment Difficulties and Loan Non-Payment Difficulties

• We observe an increase in percentage of Loan Payment Difficulties whose educational qualifications are
secondary/secondary special and a decrease in the percentage of Loan Payment Difficulties who have
completed higher education when compared with the percentages of Loan Payment Difficulties and Loan Non-
Payment Difficulties
 Final Insights

• The count of ‘Low skilled Laborers’ in ‘OCCUPATION_TYPE’ is comparatively very less and it also has maximum
% of payment difficulties- around 17%. Hence, client with occupation type as ‘Low skilled Laborers’ are the
driving factors for Loan Defaulters.

• The count of ‘Lower Secondary’ in ‘NAME_EDUCATION_TYPE’ is comparatively very less and it also has
maximum % of payment difficulties- around 11%. Hence, client with education type as ‘Lower Secondary’ are
the driving factors for Loan Defaulters.

• Banks should focus more on contract type Student ,pensioner and Businessman with housing type other than
Co-op apartment, Office apartment for successful payments.

• Banks should focus less on income type Working as they are having the greatest number of unsuccessful
payments.
 Final Insights

• Applicants living in House/Apartments has the highest number of loan application. While we see that Rented
apartment and applicants living with parents have very high percentage to default

• Get as much as clients from housing type With parents as they are having least number of unsuccessful
payments

• Also, with loan purpose Repair is having higher number of unsuccessful payments on time.
THANK YOU