Evaluation of Parkinson's Motor Symptoms in Clinical and Free-Living Conditions

Objective and continuous assessment of motor symptoms in Parkinson's disease (PD) is limited by the subjectivity and restricted temporal coverage of conventional clinical evaluations. Smartwatches enable objective motor data acquisition, although unsupervised recordings introduce additional variability. This work compares the performance of machine learning (ML) models for classifying PD motor symptoms using inertial data collected from a wrist-worn smartwatch during selected clincal tasks. Triaxial accelerometer and gyroscope signals were acquired from 24 PD patients and 16 healthy controls under supervised clinical conditions and unsupervised homebased conditions. Biomechanical features were extracted and used to train K-Nearest Neighbors, Support Vector Machine, and Random Forest classifiers for tremor, bradykinesia, and gait-related tasks. Results show that supervision level and model selection significantly affect classification performance, supporting the use of smartwatch-based ML approaches for objective PD motor assessment in real-world settings.