Hierarchical Boosting Dual-Stage Feature Reduction Ensemble Model for Parkinson's Disease Speech Data

Area of Science:

• Biomedical Engineering
• Machine Learning
• Computational Neuroscience

Background:

• Parkinson's disease (PD) diagnosis is challenging in early stages.
• Speech data analysis offers a promising avenue for early PD detection.
• Existing machine learning models struggle with speech data noise and redundancy, impacting diagnostic accuracy.

Purpose of the Study:

• To introduce a novel Hierarchical Boosting Dual-Stage Feature Reduction Ensemble Model (HBD-SFREM) for improved early Parkinson's disease diagnosis using speech data.
• To address the limitations of traditional one-sided feature reduction methods in speech analysis for PD detection.

Main Methods:

• Development of HBD-SFREM, featuring an iterative deep extraction mechanism to build a deep hierarchy instance space.
• Integration of manifold feature extraction with nearest neighbor feature preference to create a dual-stage feature reduction.
• Iterative application of the dual-stage reduction via AdaBoost to enhance instance feature quality and model generalization.

Main Results:

• HBD-SFREM demonstrated significantly improved accuracy in Parkinson's disease speech recognition across multiple datasets.
• The model's performance remained robust even with small sample datasets.
• HBD-SFREM outperformed other feature reduction and deep learning algorithms in PD speech recognition tasks.

Conclusions:

• HBD-SFREM offers a superior approach to feature reduction for Parkinson's disease detection using speech data.
• The proposed model enhances diagnostic accuracy and generalization, providing a valuable reference for future research in neurodegenerative disease identification.