Early Prediction of Parkinson's Disease (PD) Using Ensemble Classifiers

Abstract

Parkinson's Disease (PD) is a progressive neurodegenerative disorder mainly characterized by motor and non-motor symptoms. Speech symptoms represent as one of the earliest motor indications, exhibited by approximately ninety percent of PD patients. There is a need for prediction of PD based on speech symptoms using machine learning techniques in order to delay the progression of the disorder. The proposed framework performs an early prediction of PD using ensemble classifiers to the speech features dataset which has been retrieved from UCI repository. The speech features dataset consists of multiple feature subsets of 252 subjects with each subject of three speech feature instances. The highly correlated features are integrated using Principal Component Analysis (PCA) and Linear Discriminant Analysis (LDA) without any loss of information. The prediction of PD is performed using ensemble classifiers because prediction results are enhanced by combining several same or different models. The various ensemble classifiers used in the analysis of PD prediction are bagging, Adaptive Boosting (AdaBoost), Gradient Boosting Machine (GBM) and Extreme Gradient Boosting (XgBoost). The ensemble classifiers are fed with the optimal parameters procured using hyper parameters tuning process such as random search and grid search. Ensemble classifier models are evaluated based on the metrics such as Accuracy, Precision, Recall, F1 score and Support. The experimental results show that PCA performs better than LDA, grid search hyper parameters tuning process provides improved results than random search and the prediction results signifies that boosting classifier provides higher accuracy than bagging classifier.