High-Accuracy Classification of Parkinson's Disease Using Ensemble Machine Learning and Stabilometric Biomarkers.

Area of Science:

• Neurology
• Biomedical Engineering
• Data Science

Background:

• Differentiating Parkinson's disease (PD) from healthy aging is critical for effective patient management.
• Postural sway abnormalities are a key motor symptom in PD, making stabilometry a potential diagnostic tool.
• Machine learning (ML) offers advanced analytical capabilities for objective diagnostic markers.

Purpose of the Study:

• To develop and validate high-performance ML models for classifying PD patients and healthy older adults (HOAs).
• To utilize quantitative stabilometry parameters for objective PD diagnosis.

Main Methods:

• Collected stabilometry data from 26 PD patients and 37 HOAs (aged 60-80) using a force platform.
• Extracted 34 time- and frequency-domain center-of-pressure (COP) sway parameters under eyes-open and eyes-closed conditions.
• Trained and validated ensemble ML models (Random Forest, Gradient Boosting, SVM) using cross-validation and stratified train-test splits.

Main Results:

• The ensemble voting classifier achieved high accuracy (0.91) and AUC ROC (0.97) in distinguishing PD from HOAs.
• Key differentiating biomarkers identified include anteroposterior sway velocity (eyes open) and total sway path (eyes closed).
• The models demonstrated excellent discriminative power, highlighting the potential of stabilometry in PD diagnosis.

Conclusions:

• An ensemble ML approach using stabilometric features provides a highly accurate and non-invasive method for PD detection.
• This technique can potentially augment clinical assessment and monitoring of Parkinson's disease.
• Objective markers derived from postural sway analysis show promise for early and accurate PD diagnosis.