Wearable Sensors and AI Classify Parkinson's Disease Severity Accurately.

This paper details a comparative system designed for the accurate classification of Parkinson's disease (PD) severity, leveraging wearable sensor technology and artificial intelligence. The approach involves analyzing motion and tremor data captured by triaxial Inertial Measurement Unit (IMU) sensors, which record acceleration and gyroscope signals across three directions (X, Y, Z). These signals provide crucial information about the subtle motor deficits characteristic of PD, such as tremors, rigidity, and bradykinesia. The study evaluated several machine learning classification models, including SVM, Logistic Regression, KNN, Decision Tree, XGBoost, and LightGBM. While models like SVM and XGBoost showed strong performance, the Light Gradient Boosting Machine (LightGBM) consistently outperformed others, achieving an impressive 97% accuracy, precision, recall, and F1-score. This demonstrates the proposed machine learning approach's robust and effective predictive capability in identifying PD severity, offering a non-invasive and data-driven method for early and appropriate identification of the disease.