The Problem
Predicting credit default is crucial for banks. Using the AMEX dataset, I explored how to build a robust scorecard model.
My Approach
- Data Cleaning: Handling missing values using KNN imputation.
- Feature Engineering: created aggregated features based on customer transaction history.
- Modeling: Used XGBoost and CatBoost.
Results
We achieved a Gini score of 0.78, which puts this model in the top 10% of submissions.