Abstract
This paper compares and analyzes machine learning (ML) and deep learning (DL) techniques with different feature selection methods for classifying Parkinson’s disease based on a speech impairments dataset. The dataset was explored by various methods to obtain vocal features, including Mel Frequency Cepstral Coefficients (MFCCs), Wavelet Transform Features, Vocal Fold Features, and Tunable Q-factor wavelet transform (TQWT). It consists of 754 features and 756 instances. In addition, the study evaluates various classification techniques, including Support Vector Machine (SVM), Decision Tree (DT), Logistic Regression (LR), Random Forest (RF), K-Nearest Neighbor (KNN), Naive Bayes (NB), Deep Neural Network (DNN), 1D Convolution Neural Network (1D CNN). Additionally, we demonstrated the effectiveness of feature selection techniques, namely Principal Component Analysis (PCA), Minimum Redundancy Maximum Relevance (mRMR), and SelectKBest. Among these, the most notable accuracy and precision were achieved by the KNN algorithm combined with PCA, recording values of 93.43% and 93.45%, respectively. Moreover, the Random Forest (RF) algorithm achieved an outstanding recall rate of 100%