A nonlinear decision tree based classification approach to predict the Parkinson's disease using different feature sets of voice data

In the past few years, lot of researchers are working to get some breakthrough for early detection of Parkinson's disease. As the old age population is increasing at a higher rate as well as it is predicted that the old age population will increase to a much higher total by 2050, it's a become a rising concern to the developed countries because the cost due to the healthcare service of these disease is really high. Parkinson's disease (PD) belongs to the group of neurological disorder, which directly affect the bra in cells and the effect is shown in terms of movement, voice and other cognitive disabilities. Researchers are keep working on different fields such as gait analysis as well as on speech analysis to find the predictors of the Parkinson's disease. Recently machine learning based approach has been used by many researchers across the field because of its accuracy on the complex data. Machine learning based approach has been used in many cases of Parkinson's disease using gait data as well as voice data. However, so far no body ha s compared the performance metrics using different feature sets by applying non-linear based classification approach based on the voice data. So in this paper we have proposed a new approach by comparing the performance metrics with different feature sets such as original feature sets as well as Principal component Analysis based feature reduction technique for selecting the feature sets. We have used non-linear based classification approach to compare the performance metrics. We have found an accuracy of 96.83% using random forest classifiers using PCA based feature sets. This analysis will help the clinicians to differentiate the PD group from healthy group based on the voice data.