Upper Limb Motor Imagery Task Classification for EEG-based Brain Computer Interface Development

Motor imagery (MI) research in brain-computer interfaces (BCIs) for specific limb movements is an important field with practical applications. There is a necessity to create a multi-category paradigm for MI that encompasses single-limb movements and accurately interprets them, which is essential for the progression of MI-based BCIs. Moreover, categorizing brain activity among diverse individuals continues to be a substantial hurdle in the field of MI-based BCIs. Motivated by this, a new approach for classifying six different upper limb movement tasks based on electroencephalogram (EEG) signals is presented in this study. Various time-domain statistical and band-power features are obtained from the pre-processed EEG signals. These features, along with an improved ensemble classifier, achieved an overall micro-averaged accuracy of 97.50% and a mean accuracy of 99.17% in categorizing diverse MI tasks from EEG signals. The derived features were evaluated using multiple machine learning classifiers to demonstrate their capability to capture distinct information between classes. The proposed method outperforms recent approaches applied to the same dataset, indicating its potential to advance MI-based BCIs development across various applications.