Machine Learning Based Parkinson's Disease Detection from Enhanced Speech

Parkinson's Disease (PD) is a disorder that affects the brain and results in uncontrollable movements. This disorder causes difficulty with balance and coordination. This paper aims to detect Parkinson's disease using denoised speech signals by applying different machine learning algorithms. When the speech signals are taken in real time, they may be contaminated with different noises. So, perfect results may not be available. In this work, speech signals for the analysis are denoised using the Weiner filter before feature extraction. Using enhanced speech signal with 40 healthy people and 40 people with Parkinson's, 13 MFCC coefficients are extracted. These features are used for classification using Decision trees, Random Forests, Support Vector Machines (SVM), and K-Nearest Neighbors (KNN). Simulation results indicate the efficacy of Random Forest algorithm as it gives the best performance and an accuracy of 95.8%. Higher accuracy may be attempted by using multirate filters or wavelet filters for denoising as an alternative to Weiner filter.