Automatic Detection of Parkinson’s Disease using EEG Signals: A Machine Learning Approach

The electroencephalogram (EEG) is a non-invasive way of recording the spiking activities of the brain using standardized electrodes positioned on the scalp. It enables the investigation of brain activities associated with many types of impairment seen by Parkinson’s disease (PD) patients. This study aims to investigate and identify a set of features for detecting PD in its early stage using EEG signals. From the EEG signals of the OpenNeuro dataset which comprises fifteen PD patients and sixteen healthy controls (HC), statistical features, and Hjorth parameters are extracted. Kruskal–Wallis test and extra tree classifier feature selection techniques are utilized to identify the highly correlated features. The best attributes are then used to train a variety of machine learning classifiers to differentiate between HC and PD patients. The best results are achieved by the decision tree classifier for PD detection, with accuracy, precision, sensitivity, and F-1 score of 98%, 97%, 99%, and 98%, respectively.