From Raw Data to Risk Classes

A practical guide to variable categorization (binning) in credit scoring model development. Covers why categorization matters for both categorical and continuous variables, including dimensionality reduction, capturing non-linear risk patterns, outlier handling, missing value treatment, and model stability. Explains graphical monotonicity analysis using equal-frequency binning, then details supervised methods including Chi-square-based grouping and Weight of Evidence (WoE)-based grouping. Includes Python implementations for computing WoE/IV, plotting default rate curves over time, and combined bar/line plots for category analysis. Demonstrates the full workflow using variables like person_income and loan_int_rate, emphasizing that binning must be statistically sound, business-coherent, and stable across train/test/out-of-time datasets.