Classification of Medication States in Parkinson’s Disease Patients Through Gait Analysis

Parkinson’s Disease (PD) is a neurodegenerative disorder often characterised by frequent and severe gait impairments. PD medication improves symptoms but leads to motor fluctuations between “ON” (effective) and “OFF” (ineffective) states. This study aims to classify medication states (ON/OFF) in people with PD using gait analysis with a single lowerback wearable sensor. Data were collected from 33 PD patients performing a 2 -minute walk test, with and without levodopa medication. A total of 181 gait features were extracted from 2 -min walk test and categorised into spatio-temporal and signalprocessing features (statistical, digital, and complex features). Significant differences in gait parameters between ON and OFF states were observed, particularly in gyroscope spectral range, step duration asymmetry, RMS of acceleration, and harmonic ratios. Four machine learning models, were trained to classify the medication states. Combining of spatio-temporal and digital features provided the best classification performance with the Logistic Regression algorithm. It achieved an accuracy of 75% and an F1-score of 76.2% using 10 features. These findings suggest that wearable sensors, combined with machine learning, can provide an objective tool for monitoring PD medication states, potentially improving personalised treatment strategies.