Parkinson Disease Classification using Hybrid Deep Learning Approach

Abstract

A vast number of people worldwide are afflicted by Parkinson’s disease (PD), a neurological condition. For prompt action to enhance patient outcomes, Parkinson’s disease diagnosis must be made as early and accurately as possible. Hence, this work proposes a hybrid deep learning approach, Convolutional Gated Recurrent Unit (CGRU) method for classification of Parkinson’s disease using varied combinations of features. The work uses a standard dataset retrieved from the UCI Machine Learning Repository and includes voice recordings of people with and without Parkinson’s disease. The work considers the TQWT (Teager-Kaiser energy operator-based Wavelet Transform) features and combines them with Wavelet Transformed-based features and the MFCC (Mel-Frequency Cepstral Coefficients) features. The proposed Convolutional Gated Recurrent Unit (CGRU) model is trained using different combinations of the extracted features namely, only TQWT features, combining TQWT and MFCC features, combining TQWT and Wavelet transformed features, and integrating TQWT, MFCC, and Wavelet transformed features. The suggested model’s performance is assessed using accuracy as the primary metric. The experimental findings show promising classification accuracies ranging between 86% to 96% with TQWD+MFCC to be the best feature combination. The high performance indicates the effectiveness of the hybrid deep learning technique particularly when multiple feature sets are combined.

Topics & Concepts