Sparse Bayesian Learning for Multiclass Classification with application to SSVEP- BCI

Sparse Bayesian Learning (SBL) is a basic tool of machine learning. In this work, multiple linear regression models under the SBL framework (namely MultiLRM), are used for the problem of multiclass classification. As a case study we apply our method to the detection of Steady State Visual Evoked Potentials (SSVEP), a problem we encounter into the Brain Computer Interface (BCI) concept. The multiclass classification problem is decomposed into multiple regression problems. By solving these regression problems, a discriminant feature vector is learned for further processing. Furthermore by adopting the kernel trick the model is able to reduce its computational cost. To obtain the regression coef- ficients of each linear model, the Variational Bayesian framework is adopted. Extensive comparisons are carried out between the MultiLRM algorithm and several other competing methods. The experimental results demonstrate that the MultiLRM algorithm achieves better performance than the competing algorithms for SSVEP classification, especially when the number of EEG channels is small.