Agenda

Topic
Case Study Business Understanding & Objective
Case Study Approach
Data Cleaning Summary
Univariate Analysis: Categorical Columns (4 slides )
Univariate Analysis: Continuous Columns ( 1 slide )
Bivariate Analysis: Between Continuous & Categorical columns(4 slides)
Correlation
EDA Summary
Additional analysis based on IV & WOE : Categorical Columns ( 2 slides)
Additional analysis based on IV & WOE : Numerical Columns ( 2 slides )
IV & WOE conclusion and summary

1
 Case Study Approach

Understanding the case study Data Cleaning Data Preparation Exploratory Data Analysis

• A brief domain • Delete unnecessary columns • Once the data cleaning is • Identify the list of categorical
understanding of banking not required for the analysis completed , review the data columns and Continuous
specialized on the loan • Delete columns that have set again and create columns on the final data set
processing null values additional data frames as • Perform univariate analysis
• Manual view of the data file needed with subset of data on categorical and
• Delete columns that have for detailed analysis
to get initial insights single unique values continuous columns
• Review the data dictionary • Arrive at a final set of individually to find the
• Delete columns or Impute columns that are required for insights
document shared along them with valid values for
which provides description of EDA step • Perform bivariate /
missing values in columns
all the columns. Using this multivariate analysis on mix
we can arrive at only a • Check the data type of the of categorical and
subset of columns that are columns and convert them to continuous columns to arrive
needed for the objective correct data types needed insightful observations
for analysis
• Load the data file into the • Conduct correlation
pandas data frame and run • Delete any rows that are coefficient of some of the
several commands to find identified as part of outliers critical columns
the size , shape , datatype • Summarize the observations
etc.,.
2
 Data Cleaning Summary

• Delete Columns
• Identify the columns that have all NaN or max NULL values and drop those
columns
• Identify the columns that are not useful for the analysis and drop them (like IDs,
Description, title etc.,. )
• Identify the columns that have single unique value and drop the columns
• Duplicate data removal:
• Identify the duplicate records / identical rows and remove the records
• Remove outliers:
• Often cases where there will be skewed distribution of data below 25% or above
75% that affects the analysis and should be removed from the data set
• Missing values treatment :
• Identify the columns where there more than 30-40% of missing values and handle
them with appropriate approach by imputing with valid methods of mean/ medium
or mode

3
 Univariate Analysis: Categorical Columns

Grade Term
• Observation: Loan applicants who are part o GRADE • Loan applicants whose term is 36months have high chances
B,D,C are more defaulted when compared to other of defaulting
grades

4
 Univariate Analysis: Categorical Columns

Emp Length Verification status

• Applicants who have employment tenure as 10years are more
• Applicants who have not be verified have high changes for
likely to be defaulted
default
• which ideally does not make any sense because generally
with more experience the annual income will be high and • However, the verified status also is close to defaulters, so
hence minimal defaulting this required additional analysis with other columns to get
• we will have to do bi-variate analysis on this for further more insights
insights combining with other categorical or numerical values

5
 Univariate Analysis: Categorical Columns(3/3)

Home Ownership
Purpose • Applicants on RENT have most possibility of defaulting on
• Applicants who plan to clear other loans/ debts consolidation the loan. And the next immediate category is those
are more susceptible for defaulting applications who have mortgages have possibility of
defaulting

6
Univariate Analysis: Continuous Columns

Inference:
• The distribution for all the 3 columns loan_amnt,
funded_amnt, funded_amnt_inv are all almost the
same

• Majority of the loan amounts taken are within the

range of 4000 – 15000

• Majority of the annual income are ranging from 4ok –

80k

7
 Bivariate Analysis: Between Continuous & Categorical columns

Annual income & Purpose Annual income with Home Owner

• Applicants with annual income between 60-70K taken loan for
• Applicants with annual income of 60-70 k and with mortgage
home improvements tend to default in high number with next
are the ones who are defaulting maximum
immediate inline with the small business, credit card and
wedding

8
 Bivariate Analysis: Between Continuous & Categorical columns

Annual income with Term Loan amount with Home ownership

• Applicants those who had taken loan to repay in 60 months
• Applicants whose home ownership is 'MORTGAGE and
had more % of number of applicants getting charged off as
have loan of 13-14k are possible defaulters
compared to applicants who had taken loan for 36 months

9
 Bivariate Analysis: Between Continuous & Categorical columns

Loan amount with purpose Loan amount with employee length

• Borrowers seeking loan amounts between 11k and 14k for the
• Borrowers with a decade of work experience who have
purpose of small businesses tend to have a higher defaulter
borrowed between 13,000 and 14,000 dollars tend to default
rate.
on their loans.
• The subsequent group with a high likelihood of defaulter
primarily consists of those seeking loans for credit card and
debt consolidation purposes.

10
 Bivariate Analysis: Between Continuous & Categorical columns

Grade with Interest rate Loan Status with Interest Rate

• A grade, used by banks to assess borrowers, ranges from A for high • As the interest rates are increasing beyond 11% there is a trend in
ratings to F for low ratings. the defaulters, and it increases as it moves beyond 14%
• Interest rates rise as the grade shifts from A to F, signifying that
higher-graded applicants receive lower interest rates, while rates
increase with a move from A to F.
• Furthermore, with an increase in grade and interest rates, there's a
corresponding increase in the likelihood of default, particularly in
immediate concerns such as defaults in credit card and debt
consolidation.

11
 Bivariate Analysis: Between Continuous & Categorical columns

Grade with Interest rate Loan Status with Interest Rate

• A grade, used by banks to assess borrowers, ranges from A for high • As the interest rates are increasing beyond 11% there is a trend in
ratings to F for low ratings. the defaulters, and it increases as it moves beyond 14%
• Interest rates rise as the grade shifts from A to F, signifying that
higher-graded applicants receive lower interest rates, while rates
increase with a move from A to F.
• Furthermore, with an increase in grade and interest rates, there's a
corresponding increase in the likelihood of default, particularly in
immediate concerns such as defaults in credit card and debt
consolidation.

12
Correlation

Inference:
• Columns installment, funded_amnt, loan_amnt have high correlation as per the
cluster map

• Public records related fields pub_rec & pub_rec_bankrupcies and number of

accounts related fields open_acc & total_acc are correlated

• Also int_rate has a bit of an impact on installment , funded_amnt, loan_amt,

pub_rec

13
 EDA Summary
 Observation on Grades: Loan applicants falling within grades B, D, and C demonstrate higher defaulter rates compared to other grade categories.

 Term Impact on Defaulting: Loan applicants with a term of 36 months show a higher likelihood of defaulting.

 Employment Length and Defaulting: Applicants with a tenure of 10 years in employment are more prone to defaulting. However, this contradicts the typical understanding that
longer job experience correlates with higher annual income and lower default rates. Bivariate analysis is necessary to explore this in conjunction with other factors for deeper
insights.

 Verification Status and Default Probability: Applicants who have not been verified exhibit a higher chance of default. However, verified status also shows a proximity to defaulters,
necessitating further analysis with additional columns for comprehensive insights.

 Loan Purpose and Default Probability: Applicants intending to clear other loans or consolidate debts are more susceptible to defaulting.

 Home Ownership and Defaulting Probability: Those who rent their homes have the highest likelihood of defaulting on a loan, followed by applicants with mortgages.

 Annual Income, Purpose, and Defaulting: Applicants earning an annual income between $60,000 and $70,000, seeking loans for home improvements, tend to default in higher
numbers, followed closely by those seeking loans for small businesses, credit cards, and weddings.

 Annual Income and Home Ownership Impact on Defaulting: Applicants with an annual income of $60,000 to $70,000 and with mortgages are the most frequent defaulters.

 Loan Amount and Home Ownership Impact on Defaulting: Applicants owning homes with mortgages and having loans within the range of $13,000 to $14,000 show a propensity
for defaulting.

 Loan Amount and Purpose Impact on Defaulting: Borrowers seeking loan amounts between $11,000 and $14,000 for small business purposes exhibit a higher rate of defaulting,
followed by those seeking loans for credit cards and debt consolidation.

 Loan Amount and Employee Length Impact on Defaulting: Borrowers with a decade of work experience, borrowing between $13,000 and $14,000, tend to default on their loans.

 Relationship between Grade and Interest Rate: Grading, used by banks to evaluate borrowers from A to F, influences interest rates. Interest rates rise with a shift from A to F,
indicating that higher-graded applicants receive lower rates. This increase in grade and interest rates corresponds with a higher likelihood of default, particularly in credit card and
debt consolidation defaults.

 Loan Status in Relation to Interest Rates: There's an observable trend in defaulters as interest rates exceed 11%, further increasing as it surpasses 14%.
14
 Additional analysis based on IV & WoE : Categorical Columns
Graphs Plotting
IV – Informatica Value
WoE – weight of evidence
*refer next slide for summary

15
 Additional analysis based on IV & WoE: Categorical Columns
IV – Informatica Value
Summary WoE – weight of evidence
*refer previous slide for graphs

Customers with Lower Default Probability:

• Shorter loan terms associate with a decreased likelihood of default.

• Higher credit grades (above B5) correspond to reduced default probabilities, with an inverse relationship—higher grades indicate lower default chances.

• Verified income is linked to a lower likelihood of default.

• Ideal loan purposes with lower default chances include weddings, major purchases, and credit cards.

• Following closely, car loans, home improvement, and vacation purposes exhibit lower default probabilities.

• States with the most favorable customers include WY, DC, MS, DE, VT, AR, TN, TX, KS, AL, MA, WV, and PA.

Customers with Higher Default Probability:

• Longer loan terms are associated with an increased likelihood of default.

• Lower credit grades (below C1) are likely to result in defaults, with a higher chance of defaulting as the grade decreases.

• Unverified income correlates with a higher chance of default.

• Loans intended for funding small businesses represent the least favorable purposes in terms of default probability.

• States with less desirable customers in terms of default likelihood include NE, NV, SD, AK, FL, HI, MO, NM, ID, OR, and CA.

16
 Additional analysis based on IV & WoE : Numerical columns
Graphs Plotting
IV – Informatica Value
WoE – weight of evidence
*refer next slide for summary

17
 Additional analysis based on IV & WoE : Numerical columns
IV – Informatica Value
Summary WoE – weight of evidence
*refer previous slide for graphs

Good Customers with Lower Default Probability:

• The interest rate for the loan plays a significant role in default probability; lower interest rates correlate with a reduced chance of default.

• Optimal interest rate: Below 8% offers the lowest default likelihood, while below 10% stands as the next favorable option.

• Higher income levels are associated with a decreased likelihood of default.

• A lower credit utilization percentage is linked to a reduced chance of default.

Bad Customers with Higher Default Probability:

• The interest rate for the loan greatly impacts the chance of default; higher interest rates correspond to an increased probability of default.

• Loans above 14% pose a high chance of default and should be avoided.

• Lower income levels are linked to a higher probability of default.

• Higher credit utilization percentages are associated with an increased chance of default.

• A higher likelihood of default is indicated if more than 2 inquiries have been made in the last 6 months.

18
 IV & WoE conclusion and summary
IV – Informatica Value

Preferred Customer Profile: Less Preferred Customer Profile:

WoE – weight of evidence

Loan Request Details: Loan Request Details:

• Seeks a shorter-term loan duration. • Seeks a long-term loan duration.
• Holds a credit grade higher than B5. • Holds a credit grade lower than C1.
• Verification of income is confirmed. • Lacks verified income.
• Optimal loan purposes: weddings, major purchases, and credit • Seeks a loan specifically for funding a small business.
cards. Subsequently, car, home improvement, and vacation rank as • Resides in one of the following states: NE. NV, SD, AK, FL, HI, MO,
the next best options. NM, ID, OR, CA.
• Resides in one of the following states: WY, DC, MS, DE, VT, AR, TN, • Prefers a higher interest rate, exceeding 14%.
TX, KS, AL, MA, WV, PA. • Lower income status.
• Prefers a lower interest rate, ideally not exceeding 8%; otherwise, • Maintains a higher credit utilization percentage.
10% is acceptable. • Has two or more inquiries in the past 6 months.
• Higher income bracket.
• Maintains a lower credit utilization.

Key consideration for Loan Approval:

When approving loans, Income Verification (IV) and Weight of
Evidence (WOE) serve as the sole objective analysis criteria,
focusing on the outlined factors. 19