Integrated Transfer Learning Based on Group Sparse Bayesian Linear Discriminant Analysis for Error-Related Potentials Detection

Brain-computer interface is a technology that is helpful for these people with dyspraxia or strokes to obtain the ability to communicate with others or control devices again. However, due to the brain signal collected by the system has terrible quilty, the error decision is often made by the BCI system, which hinders the development of the technology. Therefore, detecting the error from the BCI system holds a great significance by error-related potential (ErrP) generated when erroneous feedback from the system is found by the subject. In this paper, we propose an integrated transfer learning based on Group Sparse Bayesian Linear Discriminant Analysis (ITL_GSBLDA) to detect ErrPs. In this way, the Group Sparse Bayesian Linear Discriminant (GSBLDA) has better performance with the help of transfer learning. The experiment has been finished with the dataset of Kaggle competition. In the experiment, sensitivity, specificity, and Area Under Curve (AUC) are used to evaluate the performance of the decoder. Finality, the results are 71.49% sensitivity, 66.49% specificity, and 0.7624 AUC when using the signal features and meta-feature. And in this condition, our decoder surpasses the 5th place in the competition.