Replication Package: Probability Calibration and Synthetic Resampling Benchmark across Healthcare Datasets

This repository provides the full replication package for our empirical benchmark evaluating probability calibration methods (Platt Scaling, Isotonic Regression) and synthetic imbalance techniques (SMOTE, Class Weighting) across six standardized healthcare datasets (Pima, ILPD, Hepatitis, Cervical Cancer, CKD, Thyroid). Key contents include:- Cleaned dataset CSV files and automated audit logs (cleaning_log.json, cleaning_summary.md).- Complete Python pipeline scripts for executing the 4,050-fit 5x5 cross-validation experimental grid.- Full out-of-fold probability predictions (~43,600 rows) and statistical significance test outputs.- CPU timing analysis metrics and plotting utilities for generating publication-ready reliability diagrams.- A consolidated archive (icic_2026_replication_package.zip) preserving the original project folder hierarchy.