Parkinson’s disease identification using machine learning based on voice disorder

Parkinson's disease is the second most prevalent late-life neurological illness after Alzheimer's (PD). It is common around the globe and mostly affects individuals over 60. Tremors, stiffness, and sluggish motion are symptoms of Parkinson's disease, and persons with PD have more severe deterioration. In recent years, neural networks have become more popular in prediction issues. This paper described a unique method for detecting Parkinson's disease using ANN (Artificial Neural Networks) and KNN (K-Nearest Neighbor), emphasizing speech problems. It includes many settings for predicting Parkinson's disease based on extracted attributes from 26 separate speech samples per individual. The LOSO (leave-one-subject-out) technique will be used to verify the results. Based on Pearson's coefficient of correlation, Mathews' coefficient of correlation, principal component analysis, and self-organizing charts, a few feature selection strategies were applied to boost the algorithms' efficiency and minimize data. Matthews' correlation coefficient-based function collection yielded the best test accuracy results, and the most important voice samples were identified. Multiple ANNs are the most effective classification approach for identifying Parkinson's disease without needing the function selection method (on raw data). Finally, a neural network with a test precision of 96.47% is completed.