Parkinson's Disease Detection Using Filter Feature Selection and a Genetic Algorithm with Ensemble Learning

Area of Science:

• Computational neuroscience
• Medical informatics
• Machine learning applications in healthcare

Background:

• Parkinson's disease (PD) is a progressive neurodegenerative disorder impacting motor and non-motor functions, significantly reducing patient quality of life.
• Accurate and early detection of PD is crucial for effective management and treatment strategies.
• Voice analysis presents a non-invasive method for potential PD detection due to characteristic vocal changes.

Purpose of the Study:

• To investigate the efficacy of filter feature selection, ensemble learning, and genetic selection in detecting Parkinson's disease using voice data.
• To compare the performance of various classification models on different datasets for PD detection.
• To evaluate the impact of feature selection and ensemble methods on the accuracy and precision of PD patient identification.

Main Methods:

• Utilized two distinct datasets comprising voice features from PD patients and healthy individuals.
• Applied filter feature selection by removing quasi-constant features.
• Tested and compared the performance of decision tree, random forest, and XGBoost classifiers.
• Implemented ensemble learning methods (voting, stacking, bagging) to enhance classification performance.
• Employed genetic selection for feature evaluation and subsequent classification.

Main Results:

• Decision tree and random forest classifiers achieved 100% accuracy on Dataset 1 after feature selection.
• Ensemble learning methods were explored to further optimize the performance of top-performing models.
• Genetic selection demonstrated high precision in identifying PD patients compared to healthy individuals in most scenarios.
• Classification performance was superior on Dataset 1 compared to Dataset 2, likely due to Dataset 2's larger feature set.

Conclusions:

• Machine learning, particularly with optimized feature selection and ensemble techniques, shows high potential for accurate Parkinson's disease detection from voice data.
• The applied methods, especially decision tree and random forest on filtered data, offer a promising avenue for non-invasive PD diagnosis.
• Further research validating these findings on diverse datasets is warranted to establish clinical utility.