Predicting early onset of Parkinson's disease and identifying the most effective features

Parkinson's disease (PD) is an extremely progressive neurodegenerative disorder affecting motor and cognitive skills to a varying degree, predominantly in the age groups of 60 or above. Although medically, insufficient dopamine and absence of cells in the midbrain region lead to its manifestation; clinical diagnosis is challenging due to limited common symptoms. In this study, time-frequency based feature measurements extracted from 1040 voice samples of PD patients and healthy individuals were analyzed using supervised machine learning techniques. Normalized data features were classified using Random Forest (RF), Gaussian Naive Bayes (GNB), Decision Tree and Logistic Regression. Prediction accuracy of Random Forest was determined to be highest at 71% for this data, followed by Logistic regression 68%, Decision tree 65%, and GNB 61%. Performance of these classifiers has been compared in terms of their accuracy, sensitivity and precision-performance metrics. The insights from our study would be key in terms of providing better treatment options and establishing a superior clinical diagnosis.