Predicting Parkinson's Disease Clinical Stages with Extreme Learning Machine on Center of Pressure Data

Parkinson's disease (PD) is a progressive disorder that affects body movement, with postural instability contributing to fall risk.This study applies machine learning to assess whether postural sway features derived from center-of-pressure (COP) data can differentiate PD stages classified by the Hoehn and Yahr (H&Y) system.COP data from 32 PD patients (mean age 65.5 years, mean disease duration 7.4 years) were collected during bipedal balancing on stable and unstable surfaces with eyes open and closed.Thirteen time-domain features were extracted, and an extreme learning machine (ELM) was trained to predict PD stages based on four metrics: accuracy, recall, precision, and F1-score.The features were evaluated across nine cases with selective removal of specific COP variables.Results showed high classification accuracy across all cases (over 80%), with Instance 3 achieving the highest accuracy (91.7%) by excluding "mean velocity" in both sway directions under stable and unstable conditions.These findings suggest that COP-based postural sway measurements can effectively indicate PD progression, with specific variables reflecting distinct physiological mechanisms in postural control.