ShichenXie/scorecardpy

What it solves

It simplifies the development of traditional credit risk scorecard models by automating repetitive and complex tasks associated with credit scoring, such as variable selection and data binning.

How it works

The library provides a suite of tools to handle the end-to-end scorecard pipeline: it filters variables based on missing rates and Information Value (IV), performs Weight of Evidence (WoE) binning to transform continuous or categorical data, scales the resulting model into a credit score, and evaluates performance using metrics like AUC and PSI.

Who it’s for

Data scientists and risk analysts working in financial services who need to build traditional credit scoring models using logistic regression.

Highlights

• Automated Binning: Includes woebin for Weight of Evidence binning and woebin_adj for manual adjustment of breaks.
• Variable Selection: Tools like var_filter and iv to identify the most predictive features.
• Performance Evaluation: Built-in functions for performance evaluation (perf_eva) and Population Stability Index (perf_psi).
• Optimized Execution: Uses pre-compiled regex and sklearn.metrics.roc_auc_score for faster processing.