Metaheuristic-driven machine learning study for early detection and classification of Parkinson's disease using...

Area of Science:

• Neurology
• Biomedical Engineering
• Artificial Intelligence

Background:

• Parkinson disease (PD) is a neurodegenerative disorder affecting millions, particularly older adults, causing motor and speech impairments.
• Current monitoring challenges necessitate noninvasive, accurate, and remote diagnostic methods for early intervention.
• Aging populations increase the demand for accessible and reliable PD detection solutions.

Purpose of the Study:

• To develop and validate an automated classification system for early Parkinson disease detection using voice recordings.
• To enhance diagnostic accuracy through advanced machine learning techniques and feature selection.
• To explore the potential of telemedicine for noninvasive PD diagnosis and patient management.

Main Methods:

• Voice recordings from 31 Parkinson disease patients and healthy subjects were analyzed.
• A Light Gradient Boosting Machine (LightGBM) classifier was employed for automated classification.
• Metaheuristic-based feature selection using the Pelican Optimization Algorithm (PAO) and hyperparameter optimization were utilized to improve model performance.

Main Results:

• The baseline LightGBM classifier achieved 95% accuracy in PD detection.
• The optimized model, incorporating PAO-based feature selection and hyperparameter tuning, reached 97% accuracy.
• The enhanced model demonstrated high sensitivity, specificity, precision, and AUC, confirming its effectiveness.

Conclusions:

• Feature selection and hyperparameter tuning significantly improve the accuracy of AI models for PD detection from voice data.
• The developed system shows promise for noninvasive, remote diagnosis of Parkinson disease, facilitating early intervention.
• This approach supports the development of telemedicine solutions to enhance the quality of life for individuals with PD.