fuzzy-augmentation-reject-inference

What problem does it solve?

Build a Through-the-Door training set by scoring rejected applicants with a PD model, duplicating them as both good and bad, and assigning PD-based sample weights to augment the training data.

Core Features & Use Cases

• Fuzzy augmentation to derive PD-based weights for rejected applicants and create dual-record augmentations.
• Combine accepted data with augmented rejections to form a weighted training set suitable for downstream predictors.
• Integrate with autogluon-tabularpredictor-fit for modeling augmented data.

Quick Start

Train a logistic regression model on accepted data to estimate PD, then use the included scripts to generate the TTD dataset and summarize results.