Dimensionality Reduction with Weighted Voting Ensemble Classification Model Using Speech Data Based Parkinson’s Disease Diagnosis

Abstract

Parkinson’s disease (PD) is a progressive neurodegenerative illness that frequently affects phonation, articulation, fluency, and prosody of speech. Speech impairment is a major sign of PD which can be employed for the earlier identification of the disease and provide proper treatment. Besides, the machine learning (ML) models can be commonly employed for PD detection and classification by the use of speech data. Since the speech data has the features of maximum data redundancy, high aliasing, and small sample sizes, dimensionality reduction (DR) techniques become essential for effective PD diagnosis. Therefore, this paper presents a new DR with weighted voting ensemble classification (DR-WVEC) model for PD diagnosis. The presented DR-WVEC model operates on different stages such as pre-processing, DR, classification, and voting process. Primarily, the speech data undergoes min–max normalization process in order to normalize the speech data. Besides, linear discriminant analysis (LDA) technique is applied for reducing the dimensionality of the features. In addition, an ensemble of two ML models, namely extreme learning machine (ELM) and Adaboost models, is employed for classification. Finally, a weighted voting-based classification process is carried out where the integration of two ML models takes place and the highest outcome is chosen as the final results. In order to assess the effective PR diagnostic outcome, an extensive set of simulations were carried out on Parkinson’s telemonitoring dataset. The obtained experimental results reported the betterment of the DR-VWEC technique over the other compared methods in terms of different aspects.