Stacked Ensemble Learning for Classification of Parkinson's Disease Using Telemonitoring Vocal Features.

Area of Science:

• Biomedical Informatics
• Machine Learning in Healthcare
• Neurodegenerative Disease Research

Background:

• Parkinson's disease (PD) is a progressive neurodegenerative disorder affecting motor and non-motor functions.
• Early and accurate PD diagnosis is vital for effective patient management.
• Machine learning (ML) offers potential for developing robust PD diagnostic tools.

Purpose of the Study:

• To develop a stacked ensemble machine learning model for accurate Parkinson's disease prediction.
• To address challenges in PD datasets, including class imbalance and feature optimization.
• To evaluate the impact of validation methodology on prediction performance.

Main Methods:

• Utilized an open-access PD dataset with 22 vocal attributes and 195 instances.
• Employed subject-wise data splitting to prevent data leakage and ensure realistic evaluation.
• Applied Synthetic Minority Oversampling Technique (SMOTE) for class imbalance and feature selection techniques.
• Developed a stacked ensemble model combining SVM, RF, KNN, and DT with logistic regression as the meta-classifier.

Main Results:

• The stacked ensemble model achieved 84.7% recording-wise and 77.8% subject-wise accuracy on unseen subjects.
• Subject-wise accuracy (77.8%) significantly outperformed individual classifiers, demonstrating model robustness.
• Feature selection using gain ratio identified optimal features for performance and interpretability.
• Rigorous subject-wise validation highlighted the critical impact of validation methodology on reported results.

Conclusions:

• Subject-wise validation and prevention of data leakage yield more realistic performance metrics for PD prediction models.
• The study emphasizes the critical importance of sound validation methodologies in healthcare ML applications.
• Findings provide a template for methodologically rigorous PD classification research, advocating for larger, multi-center validation.