Extreme Gradient Boosting Classification of Motor Imagery using Common Spatial Patterns

Brain Computer Interfaces (BCI) based on motor imagery are used to discriminate between various classes of mentally simulated movement by modelling changes in brain activity. We employ the Common Spatial Patterns (CSP) algorithm to transform electroencephalogram (EEG) signals corresponding to motor imagery which is widely used in binary motor imagery classification tasks. However, this technique is highly subjects specific and relies on identifying the subject-specific frequency bands from which highly discriminative features can be extracted for the classification of EEG signals. This paper proposes a pipeline for motor imagery classification from EEG signals based on CSP features and Extreme Gradient Boosting (XGBoost) classification that eliminates the need for frequency band selection and is robust to random noise in the recorded signals. We achieve an average Kappa score of 0.59 and an average accuracy of 69.2 percent across all nine subjects of the evaluation set in the BCI Competition 2008-Graz dataset A.