Explainable Subject-Independent Freezing-of-Gait Detection Using Wearable Sensors: Toward Real-World Digital Neurological Monitoring
Freezing of gait (FoG) is a disabling motor symptom of Parkinson's disease associated with increased fall risk, mobility impairment, and reduced independence. Continuous monitoring of FoG using wearable sensors represents an important application of digital health technologies for remote neurological care. This study presents an explainable and reproducible framework for FoG detection using tri-axial wearable accelerometry from the Daphnet benchmark dataset. Signals collected from sensors positioned at the ankle, thigh, and trunk were segmented into sliding windows and transformed into interpretable digital mobility biomarkers, including magnitude statistics, signal magnitude area, jerk-based measures, spectral entropy, and freezeindex bandpower ratios. To better reflect real-world deployment, evaluation was performed using leave-one-subject-out validation. The resulting baseline achieved an area under the precision-recall curve of 0.297 and an area under the receiver operating characteristic curve of 0.795 under subject-independent evaluation. Sensor-location ablations indicated that ankle-based sensing provided strong individual predictive performance, while multisensor fusion yielded the highest overall detection performance. These findings highlight the importance of interpretable wearable mobility biomarkers and realistic evaluation protocols for clinically meaningful digital health applications in Parkinson's disease.