Performance analysis of attributes selection and discretization of Parkinson’s disease dataset using machine learning techniques: a comprehensive approach

Prediction of Parkinson disease (PD) in an early stage is important since the disease is not curable at later stages. Many machine algorithms have been used in the currently available works to obtain a precise result. Most of the algorithms are based on random forest, Decision tree, linear regression, support vector machine (SVM), and Naïve Bayes. This paper uses four classifiers such as J48, NB-tree, multilayer perceptron neural network (MPNN), and SVM. These approaches are used to classify the Parkinson disease dataset without applying attribute selection approaches. The dataset for the work is collected from UCI Parkinson repository. The performances of the proposed four classifiers are evaluated on the original dataset, discretized dataset, and selected attributes. Based on the outcome of the study, J48 achieves high accuracy on discretized dataset. MPNN performs well with better accuracy without attribute selection and discretization on PD dataset. The results showed that SVM achieved the highest accuracy of 95.05% on the non-discretized dataset, followed by MPNN with 94.06% accuracy. J48 achieved the highest accuracy of 94.12% on the discretized dataset, followed by SVM with 93.04% accuracy. From the observation, we came to know that except MPNN all the classifiers perform well on data discretization.