Parkinson’s disease (PD) is one of the most widespread diseases that, primarily, affects the motor system of the neural central system. In fact, PD is characterized by tremors, stiffness of the muscles, imprecise gait movements, and vocal impairment. An accurate diagnosis of Parkinson’s disease is usually based on many neurological, psychological, and physical investigations despite the fact that its main symptoms cannot be easily decorrelated from other diseases. As such, many automatic diagnostic support systems based on Machine Learning approaches have been recently employed to assist the PD patients' assessment. In the current paper, a comparative analysis was performed on machine learning (ML) techniques for PD identification based on voice disorders analysis. These ML methods included the Support Vector Machine (SVM), K-Nearest-Neighbors (KNN), and Decision Tree (DT) algorithms. In addition, two feature selection techniques; mRMR and ReliefF; are used to further improve the performance of the proposed classifiers. The efficiency of the developed model has been evaluated based on accuracy, sensitivity, specificity and AUC metrics, and it is higher than existing approaches. The simulation results show that the KNN algorithm yielded the best classifier performance in term of accuracy and reached an AUC of 98.26%.
<span lang="EN-US">Brain degeneration involves several neurological troubles such as Parkinson’s disease (PD). Since this neurodegenerative disorder has no known cure, early detection has a paramount role in improving the patient’s life. Research has shown that voice disorder is one of the first symptoms detected. The application of deep learning techniques to data extracted from voice allows the production of a diagnostic support system for the Parkinson’s disease detection. In this work, we adopted the synthetic minority oversampling technique (SMOTE) technique to solve the imbalanced class problems. We performed feature selection, relying on the Chi-square feature technique to choose the most significant attributes. We opted for three deep learning classifiers, which are long-short term memory (LSTM), bidirectional LSTM (Bi-LSTM), and deep-LSTM (D-LSTM). After tuning the parameters by selecting different options, the experiment results show that the D-LSTM technique outperformed the LSTM and Bi-LSTM ones. It yielded the best score for both the imbalanced original dataset and for the balanced dataset with accuracy scores of 94.87% and 97.44%, respectively.</span>
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.