Reproducibility Materials for Hidden Risk in Low-Screening Digital Credit
Description
This repository contains the reproducibility materials supporting the empirical analysis of behavioural risk monitoring in low-screening digital credit. The archive includes analytical code, configuration files, environment specifications, tests, documentation, notebooks, and frozen analytical outputs used to reproduce the reported results. The empirical workflow covers transaction-level behavioural feature construction, predictive model comparison, temporal and user-grouped validation, probability calibration, cost-sensitive operating analysis, out-of-fold risk scoring, and user-week event analyses. The repository documents the analytical proxy-history construction, temporal information restrictions, model specifications, event definitions, sample reconciliation, and reproducibility settings. The underlying IEEE-CIS Fraud Detection source data are not redistributed in this archive. Users must obtain the original source data separately and place them in the locations specified in the repository documentation before executing the complete pipeline. The repository supports predictive and associational analyses based on the public IEEE-CIS transaction data and reconstructed analytical proxy histories. It does not contain authenticated customer identities or contractual BNPL account records. Reproducibility instructions, software requirements, expected input locations, principal outputs, and known analytical boundaries are documented in the included README and supporting files.