Enhanced Motor Imagery Classification through Channel Selection and Machine Learning Algorithms for BCI Applications

Abstract

Brain-Computer Interface (BCI) applications utilizing Electroencephalography (EEG) signals have garnered significant attention for their potential to facilitate through communication between the brain and external devices. EEG-based BCIs offer a non-invasive means to interpret neural activity, enabling a range of applications in healthcare, gaming, and cognitive neuroscience. This study explores motor imagery (MI) EEG signals classification, employing a variety of signal processing techniques as well as machine learning algorithms to increase accuracy and reliability. Using data from the BCI Competition IV dataset, the proposed methodology involves EEG band separation via Butterworth bandpass filters, channel selection through a wrapper method using K-nearest neighbors (KNN), and classification of motor imagery tasks. The study demonstrates a high classification accuracy of 98% across different motor imagery tasks, highlighting the effectiveness of the proposed approach. This method not only shows promise for BCI applications aimed at assisting individuals with motor disabilities but also for gaming and potential security applications such as user authentication. Future work will focus on further enhancing the model's accuracy and exploring its integration into diverse practical applications.