Fig. 2.
Block diagram of the signal processing and machine learning pipeline implemented in the study. The system consists of two phases. During the training phase, the features were extracted in the temporal and spatial domains from the raw EEG signals, followed by the estimation of covariate shift parameter (i.e. λ and L, smoothing constant and control limit multiplier, respectively) and a classifier is trained on the labeled examples (i.e. XTrain). In the evaluation phase, a similar signal processing method is applied initially and CSP features were monitored by the CSE and adaptation block. In the CSA block, the CSE procedure identifies the CSs and initiates adaptation by adding the kth classifier fk to the ensemble E, where k counts the number of identified CSs during the evaluation phase. Finally, the k classifier outputs from E are combined to predict the class label.