Explainable Machine Learning in Human Gait Analysis: A Study on Children With Cerebral Palsy

This work investigates the effectiveness of various machine learning (ML) methods in classifying human gait patterns associated with cerebral palsy (CP) and examines clinical relevance learned features using explainability approaches. We trained different ML models, including convolutional neural networks, self-normalizing random forests, decision trees, generated explanations for models. For deep Grad-CAM were aggregated on levels to obtain at decision, class model level. investigate which subsets 3D analysis data are particularly suitable classification CP-related patterns. The results demonstrate superiority kinematic over ground reaction force this task show that traditional approaches such as forests trees achieve better focus more clinically relevant regions compared networks. best configuration, sagittal knee ankle angles a forest, achieved accuracy 93.4 % all four CP classes (crouch gait, apparent equinus, jump true equinus). Deep networks utilized not only but also additional ones their predictions, may provide novel insights into raise new research questions. Overall, article provides application practice highlights importance promote trust understanding