An Automatic Parkinson’s Disease Classification System Using Least Square Support Vector Machine

Parkinson’s disease (PD) is a devastating neurological disease that affects millions of people throughout the world. Although there is no apparent cure for this disease, but with the help of advancement in the area of machine learning approaches, the PD can be diganosed at the early stage. Diagnosing PD at an early stage can assist in keeping the disease from progressing. This paper proposed a system to classify or predict whether the person with PD using supervised Least-squares support vector machine (LS-SVM) method. We used the UCI machine learning repository dataset, which contain the voice data acquired from 31 people of which only 8 samples are from normal people. We tested different feature selection methods for selecting optimal features from the given features in the datatset. We handled the data imbalance using Synthetic Minority Oversampling Technique (SMOTE). Finally, the Least-squares support vector machine (LS-SVM) and SVM classifiers were used to classify the data into PD and Non-PD groups. We compared the results obtained with SVM and LS-SVM with different feature selection methods. The result obtained showed that the system with the ExtraTreesClassifier feature importance method and LS-SVM classifier outperforms as compared to one which uses SVM. The classification accuracy of 98.31% was achieved using the ExtraTreesClassifier feature importance method and LS-SVM for radial kernel. The proposed system gave sensitivity = 1.0, specificity = 0.97, Precision = 0.96, Recall = 1.0, F1-score = 0.98. The proposed system is highly accurate and giving accuracy of 98.3%. The proposed system’s accuracy is superior to the cutting-edge techniques suggested in the literature.