CAD machine-learning selection stability: analysis code

Abstract

Code-only archive for the manuscript on the stability of machine-learning algorithm, feature and hyperparameter selection across random training-test splits in a coronary artery disease cohort. The archive contains selected analysis sources, recorded environment requirements, English-directory path-adaptation and provenance documentation, synthetic software tests, and a conditional split-level jackknife precision tool. The clinical models were not refitted for this release. No patient-level datasets, individual predictions, split membership tables, fitted models, clinical checkpoints, signed documents, or synthetic dataset files are included. Clinical reproduction requires institutionally authorized access to controlled inputs and retained analysis artifacts. Synthetic tests do not reproduce or independently validate the clinical findings. The software is distributed under the MIT License.