Advancing Parkinson's disease detection through multi-dimensional machine learning: a comprehensive framework using...

Area of Science:

• Biomedical Engineering
• Machine Learning
• Neurology

Background:

• Wearable movement sensors offer objective assessment of Parkinson's disease (PD) motor symptoms.
• Optimal machine learning (ML) approaches and feature sets for accurate PD detection using sensor data are not yet clearly defined.

Purpose of the Study:

• To comprehensively evaluate ML classifiers, feature contributions, and optimization techniques for PD detection using wearable movement sensor data.
• To identify the most influential features and their impact patterns for PD detection.

Main Methods:

• Compared twelve ML classifiers on motion sensor data.
• Conducted feature ablation studies across statistical, frequency-domain, dynamic, and complexity features.
• Optimized Random Forest (RF) parameters using Particle Swarm Optimization (PSO), Improved Satin Swarm Algorithm (ISSA), and Enhanced Whale Optimization Algorithm (EWOA).
• Performed SHAP value analysis to identify influential features.

Main Results:

• Random Forest achieved 86.7% accuracy, outperforming other classifiers.
• Statistical features were most significant, with complexity, dynamic, and frequency features providing complementary information.
• PSO-optimized RF reached 87.65% accuracy.
• SHAP analysis highlighted entropy-based measures and standard deviations as key features, with accelerometer and gyroscope data showing distinct influence patterns.

Conclusions:

• Ensemble ML methods effectively model the relationship between movement and PD diagnosis.
• Comprehensive feature extraction enhances PD detection accuracy.
• Findings support developing accurate, interpretable wearable systems for PD detection and management.