Reproducibility package for Leakage-Conscious Machine Learning for Supplied Injury-Risk Label Classification and Longitudinal Injury Forecasting: A Multi-Dataset Evaluation
Abstract
This record provides the reproducibility materials associated with the study “Leakage-Conscious Machine Learning for Supplied Injury-Risk Label Classification and Longitudinal Injury Forecasting: A Multi-Dataset Evaluation”. The materials include the executed and clean primary-analysis notebooks, Python scripts, fixed model and run configurations, software-version information, hyperparameter specifications, source and output manifests, machine-readable predictions, calibration outputs, cluster-bootstrap summaries, sensitivity analyses, injury-episode and alert-burden outputs, explanation-stability results, validation records, a table/figure reproducibility map, and file-level SHA-256 checksums. Release 1.0.5 additionally includes the weighted-versus-unweighted Logistic Regression sensitivity, athlete-cluster uncertainty for calibration intercept/slope, team-stability and without-team sensitivity checks, paired user-cluster uncertainty for the four-predictor personalized sensitivity, and the descriptive exact interval for 9/12 SoccerMon episode detection. SoccerMon is the primary longitudinal analysis and is evaluated using internal season-to-season temporal validation in the same athlete panel. The two supplied-label datasets are supporting methodological stress tests. The models are not presented as clinically validated or deployment-ready. Original input datasets are not redistributed. Source identifiers, retained-file characteristics, dimensions, available licensing/data-use documentation, and SHA-256 checksums are provided for provenance and verification.
