Multi-class Motor Imagery Classification in Brain Computer Interface (BCI)

A Brain-Computer Interface (BCI) represents an innovative system created to allow a communication between the human brain and a device in order to enhance the lives of disabled individuals. BCI systems decode user intentions using brain signals, detected through electroencephalography (EEG). The use of BCI with EEG gained popularity due to its non-invasiveness, high portability, and modest cost. The objective of this work is to classify different types of mental tasks involving motor imagery, such as imagining left hand, right hand, foot, and tongue movements, performed by three subjects. To achieve this, we utilized common spatial pattern (CSP) method for multiclass feature extraction. Furthermore, we compared the performance of three classifiers: k-nearest neighbors (KNN), linear discriminant analysis (LDA), and support vector machine (SVM), based on brain activity during motor imagery using the III-a dataset from BCI competition III. The final classification results revealed a notable achievement, with the highest accuracy reaching 77.03%.