Voice Biomarkers for Parkinson's Disease Prediction with Enhanced Feature Reduction Methods in Machine Learning Models
Keywords:
Parkinson’s disease, machine learning, SMOTE, accuracy, AUCAbstract
As a chronic and life-threatening condition, Parkinson’s disease (PD) causes patients to become stiff and lethargic and have
unsteady voices. There is a claim that existing PD detection approaches are useless because to their high latency and poor accuracy. To
boost the accuracy of PD detection, speech recordings were employed as biomarkers in combination with the synthetic minority
oversampling method (SMOTE). Three machine learning (ML) models namely support vector machine (SVM), K-nearest neighbors
(KNN), and random forest (RF) were employed to calculate the prediction accuracy. By implementing an unsupervised dimensional
reduction strategy, the produced model reduces unnecessary data and speeds up training and testing. Model performance is evaluated by
three factors, including accuracy, F1 score, and area under the curve (AUC) values. Experimental findings revealed that the RF model
outperforms other models with 97.4% of classification accuracy. This sort of study seeks to analyze patient speech recordings to estimate
the illness severity.