An Evaluation of Machine Learning Methods for Classifying EEG Signals Associated with Motor Imagery

Numerous problems in the robotics, medical, and advertising industries can be resolved via brain-computer interface (BCI) research. People who have impairments can utilize BCI using motor imagery (MI) techniques to operate cursors, wheelchairs, or prosthetic equipment by simply visualizing the left or right movement of their limbs. A thorough evaluation of several machine learning methods that can be used to obtain remarkable traits from electroencephalogram (EEG) signals on a motor imagery data set is made in this research. Features are retrieved from the frequency domain and the time domain. Following that, features are chosen utilizing the Random Forest and Logistic Regression techniques. Selected attributes do not exhibit much variance and produce less accurate results when supplied straight to the classifier. Using the common spatial pattern (CSP), which seeks to enhance feature variation for one class while concurrently minimizing feature variation for the other, the performance of the classifier is improved. Selected components, as features, are utilized to categorize the EEG signal using a variety of classifiers, like support vector machine (SVM), linear discriminant analysis (LDA), K-nearest neighbor (KNN), and random forest (RF). In contrast to other machine learning approaches, the SVM model had the greatest average accuracy in the frequency domain of 91.43%.