With this project, our objective is to create a fully compliant PD model to determine good and bad applicants.
My approach to solving this problem can be broadly outlined into the following steps:
- Data Cleaning and Exploratory Data Analysis. Creation of 3 new variables that capture the pattern in the data better
- Calculating Weight of Evidence to perform variable selection
- Creating our base model of predicting good and bad applicants using Logistic Regression model and creating the scorecard
- Comparing base model with ensemble models like Random Forest Classifier and XGBoost.
- Finally creating a two-cut-off point strategy for the scorecard.