A Supervised Machine Learning Approach using Different Feature Selection Techniques on Voice Datasets for Prediction of Parkinson's Disease

Among the neurological diseases, parkinson's disease is the second most common disease, which affect the old age people over the age of 65 year. It is also mentioned that the number of people affected with Parkinson's disease will increase at a higher rate until 2050, and it will be a rising concern to many 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 brain cells and the effect is shown in terms of movement, voice and other cognitive disabilities. Past few years researchers are working for detection and monitoring of the Parkinson's disease by using the speech analysis as well as the gait analysis data. Machine learning and artificial intelligence techniques are gaining popularity because these techniques are able to automate the pattern recognition process with high accuracy. However so far, no body has compared the performance metrics using different feature sets by applying nonlinear and linear classification approaches 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 genetic algorithm-based feature sets as well as Principal Component Analysis based feature reduction technique for selecting the feature sets. We have used different classification approaches to compare the performance metrics. We have found an accuracy of 97.57% using SVM with RBF by using genetic algorithm-based feature sets. This analysis will help the clinicians to differentiate the PD group from healthy group based on the voice data.