Fuzzy Augmentation Reject Inference

Purpose

Create a weighted Through-the-Door dataset by scoring rejected applicants with a KGB model, duplicating them as good and bad outcomes, and assigning sample_weight by predicted PD.

Usage

• "apply fuzzy augmentation"
• "build TTD dataset with reject inference"
• "weight rejected applicants by PD"

Instructions

1. Train a logistic regression model on accepted data to estimate PD.
2. Score rejected applicants to get PD values.
3. Create two copies of rejected rows:

- Copy A: defaultflag = 1, sampleweight = PD - Copy B: defaultflag = 0, sampleweight = 1 - PD

1. Combine accepted data with both copies of rejected data and add a source column.
2. Use ./scripts/createttddata.py to standardize the augmentation.
3. Summarize results using ./templates/ttd_summary.md.