Performance Analysis of KNN, Naïve Bayes, and Extreme Learning Machine Techniques on EEG Signals for Detection of Parkinson's Disease

The application of bio-potentials for diagnosing neurological disorders has become highly effective nowadays. This work focuses on using Electroencephalogram (EEG) to detect Parkinson's disease (PD), a significant neurological disorder. PD is considered the second most common neurological disorder in the world. Being degenerative in nature, it affects the patients progressively. The progression of the severity of this disease can be restricted by a certain limit if its symptoms can be well-treated on time. This work presents a relative analysis of the performances of three machine learning (ML) techniques in detecting PD. These are K-nearest Neighbor (KNN), Naïve Bayes and Extreme Learning Machine (ELM) techniques. Statistical-based features are evaluated from the EEG data signals of normal as well as persons with PD after preprocessing. The features evaluated are then classified using the three techniques. The results of the classifiers are evaluated with the help of some performance parameters such as accuracy, precision, sensitivity, specificity and F1 score. Based on the values of these parameters, the performances of all these techniques are compared. The comparison shows that ELM performs the best, with an accuracy of 98.84% in detecting PD. The reported methodology holds significant clinical relevance. It can offer an early, non-invasive, and objective method for diagnosing, tracking, and managing PD.