فهرست نشریات علمی دسترسی آزاد ایران

   Background and Objectives: Parkinson’s Disease (PD) is a progressive neurological disorder where timely identification and monitoring of severity are essential for effective management. Wearable technology enables continuous, at-home monitoring of PD severity, reducing the dependency on in-person evaluations. This study proposes a Multi-Class Support Vector Machine (MCSVM) with an ensemble learning approach to classify behaviors and predict PD severity.    Methods: Data was collected using wearable sensors, such as gyroscopes and accelerometers, which capture movement patterns. The classification framework integrates multiple machine learning models, including random forests, k-nearest neighbors, and decision trees, using a voting-based ensemble approach to enhance prediction reliability. MCSVM was employed to address the multi-class classification challenge, categorizing behaviors into three severity levels: mild, moderate, and severe.   Results: Experimental validation demonstrated the effectiveness of the proposed method, achieving 99.89% accuracy in activity detection and 99.95% reliability in severity predictions. The ensemble approach reduced bias and variance leading to more consistent forecasts compared to individual classifiers.    Conclusion: The results highlight the potential of wearable sensor-based systems in providing accurate, real-time PD severity assessments. By leveraging ensemble learning, the model ensures robustness against data variability, enhancing clinical applicability. The proposed system can assist physicians in managing PD by providing real-time motor-complication metrics, reducing the need for frequent clinical visits, and improving patient care.