Parkinson's Disease Classification using Gaussian Mixture Models with Relevance Feature Weights on Vocal Feature Sets

In order to perceive automatically the manifestation of dysarthria in Parkinson's disease, we propose a novel classifier which is able to categorize acoustic features and detects articulatory deficits. The proposed approach incorporates relevance feature weighting to the Gaussian mixture model in order to address the issue of high dimensionality. Besides, it learns the relevance feature weights with respect to each model along with the Gaussian mixture model parameters to deal with the specificity of the class models. In order to assess the performance of the proposed approach, we used the data collected by the department of neurology in Cerrahpasa faculty of medicine at Istanbul University. The obtained results of the Gaussian mixture models with relevance feature weights algorithm are first compared to the GMM results, and to the most recent related work. The experimental results showed the effectiveness of the proposed approach with an accuracy of 0.89 and an MCC score of 0.7.