EDA CASE STUDY

Case Study Business Problem:

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 as their advantage by becoming a defaulter. Suppose you work for a
consumer finance company which specialises in lending various types of loans to urban customers. You have to use EDA
to analyse the patterns present in the data. This will ensure that the applicants capable of repaying the loan are not
rejected.

Major problems for the scenario:

There are majorly two problems observed:

1) If a person is capable of paying loan and application is rejected, bank will be ending with Interest loss
2) If a person is not capable of paying loan and application is approved, bank will be ending with credit loss.
For the business problems mentioned above slide can be insighted from analyst percepective
and reduce the loss by giving some recommendations for bank.

This process can be achieved by EDA (Exploratory Data Analysis).

EDA Process:
1) Data Cleaning
2) Univariate Analysis
3) Bivariate Analysis.
Data Cleaning
❑ Load Application CSV File.
❑ Look for few Insights using Python commands(Describe, Info, Shape,
data types).
❑ Start Looking for Null values as they affect the Quality of Data Analysis on
the Dataset.
❑ Drop Columns Which have More than 50% of null values in the respective
columns as they do not help or show impact on our Analysis.
❑ Do not drop other columns which have Less %(Around 13) of Null values as
they can be imputed based on the type of Columns Choosing Mean, Median
and Mode.
❑ There are 41 Columns which are having more than 50% of null values which are
Dropped.
Handling Outliers for continuous Numeric
variables
❑ 'AMT_REQ_CREDIT_BUREAU_QRT’,'AMT_REQ_CREDIT_BUREAU_YEAR','CNT_FAM_MEM BERS'
are continuous numeric variables chosen form the Application CSV file that are having outliers
present in the respective columns.
❑ As per the Box Plot below ‘MEAN’ can have a impact on the actual data if
imputed. Hence it is recommended to use MEDIAN for these columns.
Identifying Outliers for continuous Numeric
variables

For ‘AMT_CREDIT’ column the For ‘CNT_CHILDREN’ column the

mean is 599076.2 where as the max mean is 0.41where as the max
amount Is 4050000. amount Is 19.
Identifying Outliers for continuous Numeric
variables

For ‘AMT_ANNUITY’ column the mean is For ‘AMT_REQ_CREDIT_BUREAU_YEAR’

27117 where as the max amount Is column the mean is 1.9where as the max
225000. amount Is 21.
Imbalance Check:

Based On Target Value 0 and 1 Based on Male & Female Gender

As you can see in the graph below the As you can see in the graph below the
data is Imbalance where Target 0 is of data is Imbalance where Female are of
91.93% and Target 1 is 65.84% and Male are of 34.16%
of 8.07%
 Univariate Analysis For Categorical Variables of Target 0 and Target 1

❑ Considering Age group Variable in target 0 and target 1 thechances

of a non-defaulter and likely to default for age group of 30-40 is high in
both the cases.

❑ For the Credit amount category the low zone is havinghigher

re-payments in target 0 and in target 1 the difficulties is for average credit
amount taken holders.

❑ No Significant difference in Housing Type, Family Status, EducationType.

❑ Also for Code Gender category around 125000 Females are more in Number as

non-Defaulters than being likely to default and we can see the male category is
high in likely to be default thanNon-Defaulters.

Note: Only limited graphs are shown in the presentation. More illustrations and graph can be found in attached python file
Correlation of numerical Columns of Target ‘0’ and Target ‘1’
❑ Highest correlation exist between AMT_CREDIT and AMT_GOODS_PRICE
there is something related to this column of Target ‘0’
❑ There is negative correlation existing between CNT_CHILDREN and
DAYS_BIRTH in Target ‘0’
❑ Least correlation exist between the DAYS_EMPLOYED and
DAYS_ID_PUBLISH in Target ‘0’
❑ Highest correlation exist between AMT_CREDIT and AMT_GOODS_PRICE
there is something related to this column of Target ‘1’
❑ There is negative correlation existing between CNT_CHILDREN and
DAYS_BIRTH of Target ‘1’ is Comparatively less with 'Target 0'
❑ Least correlation exist between the DAYS_EMPLOYED and
DAYS_ID_PUBLISH of Target ‘1’
❑ For the above correlation we could see almost the values which are correlated
with one another is same as in target 0 and 1 data frames. Also by above
imputation we could AMT_GOODS_PRICE is highly correlated with AMT_CREDIT
and also with reasonably less correlated with AMT_ANNUTITY
 Univariate Analysis For Numerical Variables of Target 0 and Target 1

❑ The client reaching credit Bureau yearly enquiries are more non-
defaulters than likely to default.
❑ The client with no children are high in number of defaulters than not
Defaulters.
❑ If the Client's income is high then the client may not default
the bank.
❑ The applicant For consumer loans is likely to be non-defaulter than
being defaulter.
❑ If the clients days employed is above 350000 than it has more number of
Non-Defaulters than likely to Default

Note: Only limited graphs are shown in the presentation. More illustrations and graph can be found in attached python file
 Bivariate Analysis on Categorical columns of 'Target 0' and 'Target
1'

❑ The female gender who doesn't own a car are more non-defaulters and its the
same the same for likely to default.
❑ The male gender who owns a House / apartment are more likely to default than
non-defaulters
❑ The core staff occupation type who owns of House/apartment are more in
number of non-Defaulters whereas the occupation type Labourers who
owns of House/apartment are high in number to likely to default

❑ The Male Gender Labourers doesn't have significant difference

between Non-Defaulting and likely to default and the female
sales staff members are more in number of Non-Defaulters than likely to
default.

Note: Only limited graphs are shown in the presentation. More illustrations and graph can be found in attached python file
 Insights Bivariate Analysis of Continuous and categorical variables
and Continuous to continuous
❑ Applicants who have high income and with no children are more
likely to default
❑ Providing a loan amount of Range 500000-2500000 to the total
income of Less 500000 are more likely to
default than non-default
❑ There are more people who haven't paid back their
loans on time with a total income of less than 500000 and are
more likely to default.
❑ The variables AMT_ANNUITY and AMT_CREDIT for both
non-defaulters and Defaulters has a strong correlation and also
has similar pattern between them

Note: Only limited graphs are shown in the presentation. More illustrations and graph can be found in attached python file
Inferences drawn performing Univariate Analysis and Bivariate analysis
of Combined data frame of Categorical Variables and Continuous
Variables.
❑ There are around 70000-80000 whose loans are approved who are likely to default and
also over 200000 applicant's loan is refused who are less likely to default this would incur
loss to the bank.
❑ There is few applicant's loan with secondary / secondary special who face difficulties to
pay loan on time than who are likely to pay on time
❑ Female Gender are more likely to not face payment difficulties then the male and hence it
is recommended to approve more loans of Female Gender than the male gender at the
same Female are High in number than who face difficulties than males
❑ Labourers are high in number of occupation type list who are likely to
default or payment difficulties
❑ The Repeater applicant has High chance of non-Defaulting and also has high chance of
defaulting when compared to new applicants
❑ No millionaire is likely to default so should not refused a application of millionaire's
application for loan and Lower Middle class people are high in number to repay the
loans
THANK YOU
-Harshad Surya Chandolu