Diagnosis of Parkinson's disease by using neural networks ensemble

Diagnosis of Parkinson's, a neurological disease, is hard specifically at its early stages. Thus, research on computer based solutions to support clinical decision making has increased recently. In this study, a new classifier method that is an ensemble of different existing classifiers is utilized to diagnose Parkinson's disease in its early stages. Underlying algorithms behind the ensemble approach are three neural networks with different learning schemes. These learning methods are Levenberg-Marquardt, Fletcher-Powell and Resilient back-propagation algorithms. When the new ensemble method is compared with the used neural network structures separately, it is observed that the new approach is superior to all existing methods. An accuracy of %96.9 is obtained with the ensemble method. The new approach proves itself as a promising method in computer-aided early diagnosis of Parkinson's disease.