Enhancing Freezing of Gait Prediction in Parkinson’s Disease Using Machine Learning and Explainable AI

Parkinson’s disease (PD) is a progressive neurode-generative disorder that affects millions of individuals worldwide, significantly impairing their quality of life through motor symptoms such as tremors, rigidity, and particularly, Freezing of Gait (FOG). FOG is characterized by transient episodes where patients temporarily lose the ability to initiate or continue walking, posing risks of falls, loss of independence, and psychological distress. This study leverages advancements in machine learning (ML) and explainable artificial intelligence (XAI) to develop and validate a predictive model for FOG. Using a publicly available dataset and employing various ML techniques, the study evaluates the performance of these models in terms of accuracy, precision, recall, and F1-score. Incorporating XAI methods enhances the interpretability of the model, making its predictions more transparent. The results indicate that a 5-second window size with a 5-second pre-FOG period and a Random Forest classifier achieved the highest performance, with an accuracy of $\mathbf{9 9. 3 4 \%}$. Interpretability analyses, including permutation importance and Partial Dependence Plots (PDP), highlight the key features influencing predictions, such as peak frequencies and statistical measures. Future work will focus on model generalizability, exploring additional datasets to improve predictive accuracy and clinical applicability.