External validation of machine learning models—registered models and adaptive sample splitting

Abstract

The proposed design and splitting approach (implemented in the Python package "AdaptiveSplit") may contribute to addressing issues of replicability, effect size inflation, and generalizability in predictive modeling studies.