Hybrid Deep Learning Model For Parkinson’s Disease Classification

Abstract

Parkinson's disease (PD) is a condition that affects many people around the world. It is important to diagnose it early and accurately to help patients. In our study, we have a new way to classify Parkinson’s using a cool technique called hybrid deep learning. We specifically used something named the Convolutional Gated Recurrent Unit (C-GRU) algorithm. We got our dataset from the UCI Machine Learning Repository. It has recordings of folks both with and without Parkinson’s. We also collected features based on Wavelet Transforms, items from TQWT (that’s the Teager-Kaiser energy operator-based Wavelet Transform), plus products known as MFCC (Mel-Frequency Cepstral Coefficients). All these features made up three different datasets. To create strong models for classifying PD, we built five C-GRU models. Each one was trained using different sets of features. Model 1 only uses TQWT features. Then, model 2 mixes TQWT & MFCC features. Model 3 combines TQWT and Wavelet Transformed features. Model 4 brings together TQWT, MFCC, & Wavelet Transformed features. Finally, model 5 takes every feature in our dataset. We measured how well our models performed by checking their accuracy. The results were exciting! Models 1, 3, and 4 achieved an impressive rate of 92.92%. Model 2 did even better with a score of 96%. Model 5 achieved an accuracy of 91.15%. These great scores show how effective our hybrid deep learning method can be, especially when we mix different sets of features together.