A Pilot Study on FoG Prediction Using Machine Learning for Rehabilitation

Walking has a significant impact on one’s quality of life. Freezing of Gait (FoG) is a typical symptom of Parkinson’s disease (PD). FoG is characterised by quick and abrupt transient falls, as a result of which the patient’s mobility is limited and their independence is lost. Thus, early detection of FoG in PD patients is necessary for diagnosis and rehabilitation. The present strategies for early detection of FoG are ineffective and have a low success rate. This study illustrates the comparative analysis of ML techniques (K Nearest Neighbors (KNN), Decision Trees, Random Forest, Support Vector Classifier (SVC), and Ada Boost Classifier), using time and statistical features to perform detection and prediction tasks on the publicly available DaphNet database. FoG prediction is highly patient dependent and achieved a peak F1 - score of 80% for one of the patients. The paper also present a combined analysis of all the patients which may aid in designing wearable sensors for detection. This system detects FoG with a precision value of about 81%.