Comparative Analysis of Machine Learning and Ensemble Learning Classifiers for Parkinson’s Disease Detection

Abstract

A neurological illness called Parkinson's disease (PD) commonly appears between the ages of 55 and 65. Moreover, a patient's entire quality of life is significantly impacted by the progressive development of motor as well as non-motor symptoms due to this disease. There is no known cure for PD, although a number of therapies have been created to assist control its symptoms. Therefore, the management of PD is a field that is expanding, and there is a need to develop a comprehensive framework for the timely detection and classification of PD. In this paper, we developed six machine learning-based and five ensemble learning-based classification models to forecast PD. The six base classifiers which are used in the current study are Support Vector Machine (SVM), Decision Trees (DTs), Random Forest (RF), K-Nearest Neighbor (KNN), Logistic Regression (LR), Naive Bayes (NB); and five ensemble classifiers named XGBoost, Gradient Boost, Bagging, CatBoost and Light Gradient Boosted Machine (LGBM) respectively, are then carefully compared. To improve the performance of the classifiers and to reduce the problem of overfitting, a feature selection method named Principal Component Analysis (PCA) and various preprocessing techniques are applied. Further, this study uses the voice samples dataset from the UCI repository having 188 PD and 64 normal patients. Overall, our findings revealed that, when compared to the other five base classifiers, the RF model offered the best classification performance with an accuracy of 82.37%, and the ensemble classifier named LGBM shows best results when compared with base as well as ensemble classifiers having an accuracy of 85.90%.