Predicting NHL Performance from Cognitive-Motor Metrics After Concussion: An Explainable AI Approach

In this work, we introduce a novel approach using cognitive-motor integration (CMI) metrics and concussion history to predict NHL participation. Specifically, we leverage data from the BrDi Test—a validated neuroscience-based assessment involving visuomotor tracing tasks under both standard and non-standard conditions. Using data from the BrDi Test—a neuroscience-based cognitive-motor integration task this research explores the link between mild traumatic brain injury (mTBI), or concussion, and NHL participation. We emphasize both model accuracy and interpretability to support human scouting decisions by comparing explainable machine learning (XAI) models with traditional black-box approaches. We frame the problem as a binary classification task and evaluate a range of machine learning models. Despite the limitations of a small, class-imbalanced dataset, a class-weighted Decision Tree achieved the best average generalization performance, offering strong interpretability alongside competitive performance. Our contributions include the first application of BrDi data to NHL prediction, a comparative analysis of XAI and black-box models for talent evaluation, and a modular ML pipeline adaptable to small-sample, high-dimensional datasets. The results support the potential value of cognitive metrics in augmenting traditional scouting decisions and provide a foundation for future research as more data becomes available.