Parkinson’s Disease Classification Using Artificial Neural Networks

Abstract

An artificial neural network multi-layer perceptron classifier was developed to make a diagnosis of Parkinson’s disease (PD) using a dataset obtained from the UCI Machine Learning Repository. The dataset consists of voice recordings from patients with PD and a control group. Multiple networks were trained to vary the number of neurons in the hidden layer between 10 and 6000 in steps of 10. The network with 280 neurons in the hidden layer had the best performance showing an accuracy of 95.23%, a precision of 96.40%, a recall of 97.10%, a specificity of 90.19%, and a F1-score of 96.75%. Artificial neural networks can be used to differentiate if a patient has PD using speech-related features. Furthermore, machine learning methods could predict other neurological diseases if the biomedical information is available.