Exploratory Data Analysis For Credit Risk Assesment
- 1. THE CREDIT RISK ANALYTICS EDA Case Study By, • Mr. Prathmesh Pise • Mr. Vishal Patil
- 2. CONTENTS Problem statement Flow Chart Importing and Cleaning1 Importing and Cleaning2 Approach Data Visualization Significant Insights
- 3. PROBLEM STATEMENT: 1. Aim is to identify patterns which indicate if a client had difficulty paying their installments which will help the bank in taking following actions: • Denying the loan • Reducing the amount of loan • Lending (to risky applicants) at a higher interest rate, etc. 2. Identifying the co-relation between dependent variables with target variable 3. To ensure that the consumers capable of repaying the loan are not rejected
- 5. 1. Imported pandas, matplotlib and seaborn library for loading the data and data visualization 2. Target variable is flag variable weather a clients pays instalments on time or not 3. Two data frames were created from csv files namely, • Application data- Contains all the information of the client at the time of application • Previous application data - contains information about the client’s previous loan data 4. Dropped unnecessary columns like the one belonging to client’s house dimensions 5. Achieved 40% memory usage reduction by changing the data types of categorical variables from object to category. IMPORTING AND CLEANING1:
- 6. IMPORTING AND CLEANING2: 1. Imported required data set for previous application data set: • Previous application data set as previous_app 2. Cleaned the data by removing columns that were less significant for analysis and were prone to containing erroneous data, namely, • WEEKDAY_APPR_PROCESS_START • HOUR_APPR_PROCESS_START, etc. 3. Achieved 40% memory usage reduction by changing the data types of categorical variables from object to category and dropping unnecessary columns
- 7. HANDLING DATA AND MISSING VALUES: 1. Checked for null values in application_data and found that: • OWN_CAR_AGE had 65.99%, OCCUPATION_TYPE had 31.35% and EXT_SOURCE_1 had 56.38% missing values • Hence decided to drop these columns 2. We also checked for null values in previous_app and found that: • RATE_INTEREST_PRIMARY had 99.64% • RATE_INTEREST_PRIVILEGED had 99.64% had of Null values • Hence we dropped them 3. The external source data had some missing values , We impute them to zero as the External agencies have not provided score for these customers meaning the client's account was not prone to be a defaulter. Hence score was assumed as zero. 4. Took average of EXT_SOURCE_1, EXT_SOURCE_2, EXT_SOURCE_3 columns creating ext_sources column. 5. In previous_app, NAME_TYPE_SUITE had 49% missing values and does not affect whether the client will default or not. Hence, we drop this column.
- 8. 6. Defined a function null_percentage to calculate null values in the columns from both the data sets. 7. Since data is imbalanced we have taken proportion of all the categories to analyse the data and have used stacked bar plots as it enhances our understanding. 8. Defined a function called stacker this function compares a categorical column with our Target variable, it considers data imbalance and converts each category into percentages and plots the stacked chart with their proportion. 9. Merged previous_app data set with application data set, to compare it with our Target variable.
- 9. DATA VISUALIZATION • Univariate analysis on following variables, 1. Target 2. Income 3. Children count • Bi-variate analysis on Target variable against the following, 1. Gender & age 2. Contract type 3. Average external score 4. Income & occupation type 5. Education type etc • Multi-variate analysis on Target variable against the following, 1. Income and education type 2. Income and previous application status
- 10. TARGET V/S GENDER Inference: • The percentage of Males that pay late installments is more than that of females. • The percentage of Females paying on time is more than that of males.
- 11. TARGET V/S CONTRACT TYPE Inference: • The clients with Cash loans tend to pay late as compared to the clients with Revolving loans.
- 12. TARGET V/S CAR Inference: • Percentage of people with No-Car and paying late installments are slightly more than that of people with Car
- 13. TARGET V/S AVG_EXT_SCORE Inference: • 50% client population who delay their installment payments have a low average external score, and it ranges from 0.2-0.4 approximately. • The clients who pay their installments on time have a moderate average score ranging from 0.3-0.5 approximately. • There are some clients who have received a very high score and they delay their installments.
- 14. TARGET V/S AMT INCOME Inference: • The clients with income less than 2 lakhs pa pay late installments among these classes. • The clients with income more that 6 lakhs pa i.e. Rich class is more likely to pay on time than other classes.
- 15. TARGET V/S INCOME TYPE Inference: • Amongst all the Income types, the Others(Maternity leaves, Students, Unemployed clients, etc.) are the one who tend to pay late installments. • The Businessman income types do not pay late installments. • The working class also have a higher percentage of people in late paying installments which is 10%.
- 16. TARGET V/S FAMILY STATUS Inference: • The clients who are Single/not married and the Civil marriage class tend to pay late installments.
- 17. TARGET V/S HOUSING TYPE Inference: • The clients who live in rented apartments and with parents tend to pay late installments. • The clients who stay in office apartments pay on time installments.
- 18. Inference: • The people who do not provide the Document2 tend to pay late installments. Hence it is advisable to make this document mandatory. TARGET V/S DOCUMENT 2
- 19. Inference: • The people who provide mobile number tend to pay installments on time. • Hence it is advisable to collect mobile number of the clients. TARGET V/S CLIENTS PROVIDING MOBILE NUMBERS
- 20. TARGET V/S AGE Inference: • The clients with age below 25 tend to pay late installments. • The clients with age of 65 and above pay the installments on time. • The possible reason is that clients below age 25 are less financially stable as compared to those above 65.
- 21. TARGET V/S OCCUPATION TYPE Inference: • Low skill laborers , Waiters/barmen staff , security staff , cooking , cleaning staff , drivers, Laborers tend to pay late installments. • Most of the accountants, High skill tech staff and HR-staffs pay the installments on time. • The obvious reason being that they represent the sectors with higher salary.
- 22. TARGET V/S CNT_CHILDREN Inference: • The clients who have count of children greater than 5 tend to pay late installments. • Most of the clients with count of children of 2 or 3 pay installments on time.
- 23. TARGET V/S NAME_EDUCATION_TYPE Inference: • The clients with academic degree pay installments on time. • The clients with lower secondary education pay late installments.
- 24. MULTIVARIATE ANALYSIS ON NUMERIC VARIABLES Inference: • A positive high co-relation is seen between good's price and amount credit • A positive high co-relation is seen between annuity amount and amount credit • A positive high co-relation is seen between annuity amount and good's price
- 25. PROPORTIONS OF CLIENTS BASED ON PREVIOUS APPLICATION STATUS Inference: • Out of the total loan applications only 63% were Approved. • 17% were Refused loan and 19% applications were cancelled by the clients.
- 26. HANDLING OUTLIERS Inference: • Outliers were observed in the annual income variable. • 99% clients had their income less than 4.75 LPA • Hence for analyzing the annual income, the analysis was limited to clients with annual income less than 4.75
- 27. TARGET V/S INCOME V/S EDUCATION TYPE Inference: • The clients with Education type as academic degree and income in range of 3-3.6 Lakhs pay late installments as compared to those with low income
- 28. TARGET V/S NAME_CASH_LOAN_PURPOSE Inference: • The clients who previously took loan for the payments on other loan pay late installments. • Following them ,are the clients with Home/Office/Land Loan and personal household expenses, they pay late installments
- 29. TARGET V/S INCOME V/S PREVIOUS APPLICATION STATUS Inference: • Clients who took loan for Business Development and annual income above 2.6 LPA pay late instalments.
- 30. TARGET V/S PREVIOUS LOAN STATUS Inference: • The clients for whom the previous loan was Refused , pay the installments late
- 31. KEY INSIGHTS • Following are the strong indicators of default 1. NAME_HOUSING_TYPE : Clients living in rented apartments 2. NAME_FAMILY_STATUS : Clients belonging to Civil marriage and those who are single/married 3. NAME_INCOME_TYPE : Maternity leave , students, Unemployed clients 4. FLAG_DOCUMENT_2 : The clients who do not provide document 2 5. FLAG_MOBIL : The clients who do not provide mobile number 6. OCCUPATION_TYPE : Low skill, Laborer, Waiters, Barmen, Security staff 7. CNT_CHILDREN : Positive co-relation between number of children with the chance of client being a defaulter 8. NAME_EDUCATION_TYPE : Clients with lower secondary and secondary/ secondary special and incomplete higher 9. EDUCATION_TYPE : Clients with academic degree and annual income between 3-3.6 lakhs 10. CASH_LOAN_PURPOSE : Clients with previous loan purpose as payment on other loans • Following clients should be targeted 1. CODE_GENDER : Females 2. NAME_CONTRACT_TYPE : Clients with revolving loans 3. FLAG_CODE_CAR : Clients with car 4. AVG_EXT_SCORE : Clients with moderate external score 5. AMT_INCOME_TOTAL : Clients with annual income greater than 6 lakhs 6. NAME_INCOME_TYPE : The businessmen and pensioners 7. FLAG_MOBIL :Clients who provide mobile number 8. DAYS_BIRTH :Clients with age of 65 and above 9. OCCUPATION_TYPE : accountants, High skill tech staff and HR-staffs pay the installments on time 10. NAME_EDUCATION_TYPE : Clients with academic degree
- 32. CONCLUSION • Based on the inferences obtained, a credit score can be set • Variables which contributes towards the chances of a client being a defaulter will be rated a low score • The variables contributing towards the chances of a client paying the installments on time, will be rated with high credit scores • Based on the final credit score, bank can take following decision, 1. Grant loan to clients with healthy overall credit score 2. Grant loan at higher interest rates to clients with comparatively low credit scores 3. Reject loan for clients with extremely low credit score
- 33. THANK YOU!