Early Prediction of Freezing of Gait in Parkinson’s Disease Using Wearable Sensors and Machine Learning

Abstract

Freezing of Gait (FoG) is a disabling motor symptom of Parkinson’s Disease (PD) that causes loss of mobility and increases fall risk. Most wearable-based FoG systems detect freezing at or after onset, limiting their preventive use. This study presents a machine learning framework to predict FoG several seconds before onset. The publicly available Daphnet dataset was used, containing tri-axial accelerometer data segmented with sliding windows and processed using domain-specific feature engineering. Several models were tested under a Leave-One-Subject-Out (LOSO) cross-validation protocol, including XGBoost, CatBoost, Random Forest, and deep recurrent networks. XGBoost achieved the best results (Recall = 90.03%, Precision = 76.76%, F1 = 82.14%). SHAP analysis confirmed physiologically relevant predictors such as gait rhythmicity, energy distribution, and freeze index. The framework can support real-time alert systems to help prevent FoG-related falls in PD. Future work will include hardware implementation, clinical testing, and development of neuromorphic variants for wearable devices.