Parkinson’s Disease Diagnosis: Detecting the Effect of Attributes Selection and Discretization of Parkinson’s Disease Dataset on the Performance of Classifier Algorithms

摘要

Precise detection of PD is important in its early stages. Precise result can be achieved through data mining, classification techniques such as Naive Bayes, support vector machine (SVM), multilayer perceptron neural network (MLP) and decision tree. In this paper, four types of classifiers based on Naive Bayes, SVM, MLP neural network, and decision tree (j48) are used to classify the PD dataset and the performances of these classifiers are examined when they are implemented upon the actual PD dataset, discretized PD dataset, and selected set of attributes from PD dataset. The dataset used in this study comprises a range of voice signals from 31 people: 23 with PD and 8 healthy people. The result shows that Naive Bayes and decision tree (j48) yield better accuracy when performed upon the discretized PD dataset with cross-validation test mode without applying any attributes selection algorithms. SVM gives high accuracy of 70% for training and 30% for the test when implemented on a discretized PD dataset and a splitting dataset. The MLP neural network gives the highest accuracy when used to classify actual PD dataset without discretization, attribute selection, or changing test mode.