Efficient Model for Prediction of Parkinson's Disease Using Machine Learning Algorithms with Hybrid Feature Selection Methods

In recent years, The development of telemonitoring and tel-diagnosis tools for assessing and monitoring Parkinson's disease (PD) has assumed growing significance. In this study, a Parkinson's disease prediction model is put forth. The monotony and distortion of speech are two of the most apparent signs of PD. Approaches based on artificial intelligence can assist specialists and medical professionals in automatically identifying these disorders. This study proposed an effective method for detecting PD from features incorporating voice recordings of subjects who have already been diagnosed. The main objective of this paper is to analyze the effect of an imbalanced dataset and feature selection. We have used SMOTE technique to balance the imbalanced data on the efficiency of the voice-based PD detection system and use a learning-based classifier for this work. Feature selection using LassoCV, Recursive Feature Elimination (RFE) and Hybrid selection techniques. Various machine learning classifiers, including Random-Forest (RF), Logistic Regression (LR), Decision-Tree (DT), Gaussian NB (GB), SGD Classifier (SGDC), Nearest-Neighbors (NN), Support Vector Machine (SVM), Adboost (Adb), MLPClassifier(MLP) and BaggingClassifier(BC), have been Classified and explored for PD detection purposes. All other models were outperformed by the Bagging classifier results with Hybrid feature selection, which had high Accuracy (A) of 91.6%, Precision (P) of 91.8%, Recall (R) of 91.6%, and F1- score (F) of 91.7% and CVS of 97.4%. The suggested method results show that the Hybrid features selection approach with Bagging Classifier (BC) successfully enhances the overall performance of the Parkinson's disease detection model.