Robust attribute selection to improve the Parkinson's disease classification: a hybrid approach

Abstract

Parkinson’s disease (PD) is known to be a neurodegenerative syndrome that progresses chronically. As a result of the damage or death of brain neurons that generate dopamine patients tend to face difficulty when performing simple everyday tasks like walking, writing, or speaking. The main contribution of this work presents a hybrid method for improving predicting PD. This methodology has been obtained by means of testing a number of different combinations of classification algorithms and approaches for selecting attributes. A total of three attributes selection methods (correlation, information gain, and variance threshold) and three classifiers (decision trees (DT), naive bayes (NB), and support vector machine (SVM)) have been adopted. The speech data set provided by University of California-Irvine (UCI) machine learning (ML) repository is adopted to analyze the performance of different combinations. The combination of information gain and DT classifier achieved the best performance rather than other combination methods, reaching a classification accuracy of (97.43%). Finally, an additional comparison of the performance analysis with the results of previous studies was made and it was found that the proposed methodology proved to outperform the results of other studies conducted in this field.