Acoustic measures for voices in the classification of Parkinson's Disease

The ART Family Networks are neural networks based on the Adaptive Resonance Theory that has as a characteristic the resonance between the input data and the cluster center as the data was classified. In particular, self-expanding ART Neural Networks characterized by the ability, even with little data, to perform the data classification and to expand according to the inclusion of the data. This work uses the Fuzzy ART Self-expanding neural network for the diagnosis of Parkinson's Disease. For this, it uses the Parkinson Speech Dataset with Multiple Types of Sound Recordings database, available in the UCI Machine Learning Repository repository. This database is composed of data from tested individuals, based on the characteristics extracted from 26 different voice samples per individual. The problem is to classify patients as healthy or with Parkinson's disease. The correlation coefficient used to select which voice resources were most relevant. After implementation using the resources that showed the strongest positive correlation, an accuracy of 98.56% obtained with a Mattews Correlation Coefficient of 0.9716 using the 10-fold cross-validation method.