Predictive Customer Default Logistic Regression
Budget: ₹600 – ₹1,500 INR
IHC - Data Analytics and Reporting: Industry Assignment
Build a logistic regression model to predict the likelihood of a customer defaulting in next 12 months.
The project involves following steps:
1. Univariate analysis to assess the data quality and summarization of the information available to
build a logistic regression model.
2. Winsorization and missing imputation on the independent variables.
3. Bivariate analysis to assess the linear relationship between the dependent and independent
variables.
4. Splitting the data into development and validation dataset and building a logistic regression
model.
5. Assessing the model performance based on the development and validation data.
Please answer the following questions based on the data set provided:
Question #
Question 1
How many rows and columns are there in the data?
Question 2
How many defaults are there in the data (Dependent variable: 'default payment next month')?
Question 3
What is the default rate in the data?
Question 4
Apply following data exclusions to prepare the model development data and compute the new default
rate: ['BILL_AMT6']>0 and ['Limit']>0 and ['Debt']>31. How many remaining rows and columns are there
in the data?
Question 5
What is the default rate in the model development data?
IHC - Data Analytics and Reporting: Industry Assignment
Question 6
Create one derived variable named payment ratio in the model development data and compute
minimum, maximum and mean of the derived variable. Derived variable logic:
'PAY_AMT6' / 'BILL_AMT6'
Question 7
Create one rescaled variable credit limit (divided by 1000) in the model development data and compute
minimum, maximum and mean of before and after of the rescaled variable. Use variable 'Limit'.
Question 8
Perform bivariate analysis on the continuous variable payment ratio (Payratio, created in question 6).
Divide the variable in the five equal bins and compute default rate in each bin.
Question 9
Perform bivariate analysis on the continuous variable credit limit (LIMIT_BAL1, created in question 7).
Divide the variable in the five equal bins and compute default rate in each bin.
Question 10
Perform bivariate analysis on the continuous variable total debt (Variable name: 'Debt'). Divide the
variable in the five equal bins and compute default rate in each bin.
Question 11
Answer following questions in 'Yes' or 'No':
a) Does the default rate monotonically decrease (based on question 8 result) as the payment ratio
increases?
b) Does the default rate monotonically decrease (based on question 9 result) as the credit limit
increases?
c) Does the default rate monotonically increase (based on question 10 result) as the debt increases?
IHC - Data Analytics and Reporting: Industry Assignment
Question 12
Compute the 1st and 99th percentile value of the variables payment ratio (Created in question 6), credit
limit (Created in question 7) and total debt (Variable name: Debt)
Question 13
Create new variables for payment ratio (Created in question 6), credit limit (Created in question 7) and
total debt (Variable name: Debt) by capping the variables at 99th percentile values. Compare the pre
and post capped variables mean. Use rescaled variable for credit limit.
Question 14
Create new variable by the name of intercept and assign a constant value of 1. Compute the mean of
this new variable.
Question 15
Create train (Insample data) and test (out of sample) data with random seed as '0'. Test data should
have 20% of the model development data records. The 'Y' variable will be 'default payment next month'
and 'X variables will be ('Debt', 'Credit limit', 'intercept' and 'Payment ratio'.
How many records are there in train and test data?
Hint: Use train test split from sklearn
Question 16
Create new train ( New insample data) and new test (New out of sample) data with random seed as '0'
based on the capped variables created in question 13. Test data should have 20% of the model
development data records. The 'Y' variable will be 'default payment next month' and 'X variables will be
('Debt', 'Credit limit', 'intercept' and 'Payment ratio'.
How many records are there in train and test data?
Hint: Use train_test_split from sklearn
Question 17
Using the train data fit the logistic model and print the model coefficient
Hint: Use statsmodels.api
Question 18
Using the new train data (Created in question 16) fit the logistic model and print the model coefficient
Hint: Use statsmodels.api
IHC - Data Analytics and Reporting: Industry Assignment
Question 19
Based on the estimates obtained using the logistic regression using the train data, answer following
questions in 'Yes' or 'No':
a) Does the estimate sign for payment ratio align with the observed trend in results for question 8?
b) Does the estimate sign for credit limit align with the observed trend in results for question 9?
c) Does the estimate sign for debt align with the observed trend in results for question 10?
d) Are all three variables statistically significant?
Question 20
Based on the estimates obtained using the logistic regression using the new train data (created in
question 16), answer following questions in 'Yes' or 'No':
a) Does the estimate sign for payment ratio align with the observed trend in results for question 8?
b) Does the estimate sign for credit limit align with the observed trend in results for question 9?
c) Does the estimate sign for debt align with the observed trend in results for question 10?
d) Are all three variables statistically significant?
Question 21
Using the train data, fit the logistic model and print the AUC values for train and test data.
Hint: Use roc_auc_score from sklearn
Question 22
Using the new train data, (created in question 16) fit the logistic model and print the AUC values for new
train and new test data.
Hint: Use Logistic Regression from sklearn
Question 23
Does the AUC value change by using pre-cap values (Created in question 21) and post-cap values
(Created in question 22)?
Question 24
Score the model generated in question 17 on the train data and print following:
a) Minimum probability
b) Maximum probability
c) Total predicted defaults
d) Predicted default rated) Are all three variables statistically significant?
Question 25
Score the model generated in question 18 on the new train data and print following:
IHC - Data Analytics and Reporting: Industry Assignment
a) Minimum probability
b) Maximum probability
c) Total predicted defaults
d) Predicted default rate
Build a logistic regression model to predict the likelihood of a customer defaulting in next 12 months.
The project involves following steps:
1. Univariate analysis to assess the data quality and summarization of the information available to
build a logistic regression model.
2. Winsorization and missing imputation on the independent variables.
3. Bivariate analysis to assess the linear relationship between the dependent and independent
variables.
4. Splitting the data into development and validation dataset and building a logistic regression
model.
5. Assessing the model performance based on the development and validation data.
Please answer the following questions based on the data set provided:
Question #
Question 1
How many rows and columns are there in the data?
Question 2
How many defaults are there in the data (Dependent variable: 'default payment next month')?
Question 3
What is the default rate in the data?
Question 4
Apply following data exclusions to prepare the model development data and compute the new default
rate: ['BILL_AMT6']>0 and ['Limit']>0 and ['Debt']>31. How many remaining rows and columns are there
in the data?
Question 5
What is the default rate in the model development data?
IHC - Data Analytics and Reporting: Industry Assignment
Question 6
Create one derived variable named payment ratio in the model development data and compute
minimum, maximum and mean of the derived variable. Derived variable logic:
'PAY_AMT6' / 'BILL_AMT6'
Question 7
Create one rescaled variable credit limit (divided by 1000) in the model development data and compute
minimum, maximum and mean of before and after of the rescaled variable. Use variable 'Limit'.
Question 8
Perform bivariate analysis on the continuous variable payment ratio (Payratio, created in question 6).
Divide the variable in the five equal bins and compute default rate in each bin.
Question 9
Perform bivariate analysis on the continuous variable credit limit (LIMIT_BAL1, created in question 7).
Divide the variable in the five equal bins and compute default rate in each bin.
Question 10
Perform bivariate analysis on the continuous variable total debt (Variable name: 'Debt'). Divide the
variable in the five equal bins and compute default rate in each bin.
Question 11
Answer following questions in 'Yes' or 'No':
a) Does the default rate monotonically decrease (based on question 8 result) as the payment ratio
increases?
b) Does the default rate monotonically decrease (based on question 9 result) as the credit limit
increases?
c) Does the default rate monotonically increase (based on question 10 result) as the debt increases?
IHC - Data Analytics and Reporting: Industry Assignment
Question 12
Compute the 1st and 99th percentile value of the variables payment ratio (Created in question 6), credit
limit (Created in question 7) and total debt (Variable name: Debt)
Question 13
Create new variables for payment ratio (Created in question 6), credit limit (Created in question 7) and
total debt (Variable name: Debt) by capping the variables at 99th percentile values. Compare the pre
and post capped variables mean. Use rescaled variable for credit limit.
Question 14
Create new variable by the name of intercept and assign a constant value of 1. Compute the mean of
this new variable.
Question 15
Create train (Insample data) and test (out of sample) data with random seed as '0'. Test data should
have 20% of the model development data records. The 'Y' variable will be 'default payment next month'
and 'X variables will be ('Debt', 'Credit limit', 'intercept' and 'Payment ratio'.
How many records are there in train and test data?
Hint: Use train test split from sklearn
Question 16
Create new train ( New insample data) and new test (New out of sample) data with random seed as '0'
based on the capped variables created in question 13. Test data should have 20% of the model
development data records. The 'Y' variable will be 'default payment next month' and 'X variables will be
('Debt', 'Credit limit', 'intercept' and 'Payment ratio'.
How many records are there in train and test data?
Hint: Use train_test_split from sklearn
Question 17
Using the train data fit the logistic model and print the model coefficient
Hint: Use statsmodels.api
Question 18
Using the new train data (Created in question 16) fit the logistic model and print the model coefficient
Hint: Use statsmodels.api
IHC - Data Analytics and Reporting: Industry Assignment
Question 19
Based on the estimates obtained using the logistic regression using the train data, answer following
questions in 'Yes' or 'No':
a) Does the estimate sign for payment ratio align with the observed trend in results for question 8?
b) Does the estimate sign for credit limit align with the observed trend in results for question 9?
c) Does the estimate sign for debt align with the observed trend in results for question 10?
d) Are all three variables statistically significant?
Question 20
Based on the estimates obtained using the logistic regression using the new train data (created in
question 16), answer following questions in 'Yes' or 'No':
a) Does the estimate sign for payment ratio align with the observed trend in results for question 8?
b) Does the estimate sign for credit limit align with the observed trend in results for question 9?
c) Does the estimate sign for debt align with the observed trend in results for question 10?
d) Are all three variables statistically significant?
Question 21
Using the train data, fit the logistic model and print the AUC values for train and test data.
Hint: Use roc_auc_score from sklearn
Question 22
Using the new train data, (created in question 16) fit the logistic model and print the AUC values for new
train and new test data.
Hint: Use Logistic Regression from sklearn
Question 23
Does the AUC value change by using pre-cap values (Created in question 21) and post-cap values
(Created in question 22)?
Question 24
Score the model generated in question 17 on the train data and print following:
a) Minimum probability
b) Maximum probability
c) Total predicted defaults
d) Predicted default rated) Are all three variables statistically significant?
Question 25
Score the model generated in question 18 on the new train data and print following:
IHC - Data Analytics and Reporting: Industry Assignment
a) Minimum probability
b) Maximum probability
c) Total predicted defaults
d) Predicted default rate