A mixed classification approach for the prediction of Parkinson's disease using nonlinear feature selection technique based on the voice recording

In recent years, the people affected with Parkinson's disease (PD) are increasing with the increase in the old age population worldwide. PD affects 2–3% of the population over the age of 65 years. As the diseases progresses it produces different abnormalities in the spinal cords and brain cells that direct affect the gait, speech, and memory. Some of the recent works pointed out that artificial intelligence technique has been successfully applied to assess the disease at different stage using the gait features as well as speech related features. So in this paper an attempt has been made to distinguish PD group from the healthy control group based on voice recordings with selected features and different classification techniques such as linear classifiers, nonlinear classifiers and Probabilistic classifiers. We have used recursive feature elimination algorithm (RFE) for selection of important features. We have implemented above mentioned classification technique and found an accuracy of 97.37%, and sensitivity of 100% with linear classifier (SVM) compared with the other classifier. We have also compare the other performance metrics such as sensitivity, specificity, positive predictive value, and negative predictive by implementing the classification techniques. This analysis helps the medical practitioner to distinguish PD from healthy group by using voice recordings.