Implementation and Evaluation of Learning Classifiers in Detecting Parkinson's Disease Using Extensive Speech Parameters

Abstract

The adverse effects of neurodegenerative diseases are aimed to be reduced if not totally diminished. Parkinson's Disease (PD), a type of neurodegenerative disease, has been a trend in research and medicine with regards to its classification and early detection. There is a count on the symptoms experienced by PD patients such as tremors, rigidity, and slowness, but the majority of these patients have an impairment in speech; thus, considering voice attributes as an outstanding feature. Using extensive voice parameters including but not limited to Mel Frequency Cepstral Coefficients (MFCC) and Tunable Q-Factor Wavelet Transform (TQWT) based features, this study does not only focus on one learning machine - which is the usual subject of related literature, but on evaluating the generalization performance of 7 classification systems including their variants. This will provide a summative report on their accuracies so that researchers can proceed to higher levels of studies. As a result, the best learning classifier utilizing the data set acquired is optimized k-NN with 95.6% accuracy. This is achieved in a 10-fold cross-validation configuration.