A Comparative Analysis of Feature Selection Algorithms for Parkinson's Disease Classification

Abstract

Parkinson's disease is a progressive neurological disorder that adversely affects the quality of life of an individual and is hard to diagnose and treat at its initial stages. It helps in proper diagnosis and targeted treatment by classifying the acoustic features extracted from people suffering Parkinson’s disease and healthy subjects responsible for this disease. Conventional classification techniques have considerable challenges with the low sample size and high dimensionality of acoustic features data of Parkinson's disease. Techniques for feature selection have been developed as efficient tools that may eliminate redundancy and discover the most relevant features. The purpose of this paper is to analyze three types of feature selection techniques, namely, filters, wrappers, and embedded methods, in order to identify the important acoustic features. These features are utilized to train and test the models for classification, SVM, AdaBoost Classifier, Decision Tree Classifier, Random Forest Classifier, Logistic Regression and XGBoost. For all these classification models, by this feature selection process, their precision improved by quite a margin even when features have been minimized. Of all these models, Random Forest and XGBoost Classifier has high precision and is very reliable. AdaBoost Classifiers also demonstrate good performance, showing that they may be very promising for dependable classification in such datasets. The present study determines key acoustic features that are crucial in the course of Parkinson's disease and brings forth new information on the molecular mechanisms of the disease. Integration of feature selection methods with machine learning algorithms helps to enhance diagnostic accuracy and optimizes computational efficiency.