Some traditional and nontraditional features are analyzed for the parkinson disease classification. Various classifiers are used for this purpose. Parkinson voice has tremor in it and it is visible in jitter and shimmer values.Jitter values are higher in parkinson subjects than the healthy one. Same results are observed in case of shimmer.The harmonic to noise ratio values are high for the healthy one. The non-traditional measures show appreciable differentiation between the two classes. All the three features DFA, Spread1 and PPE have higher values for parkinson subjects than the healthy one. The tansig transfer function neural network
gives good accuracy and sensitivity than logsig and purelin neural networks, but the later two have 100% specificity. SVM gives the best results. In SVM sensitivity is 94.5%, specificity is 100% and overall accuracy is 95.9%. In this paper, the feature data set of healthy and parkinson subject was taken. The features were selected and classified with various classifiers. The SVM was able to classify with an accuracy of 95.9%. More advanced algorithms arebeing considered for better selection and classification of recorded samples.