Batch mode MS-based and entropy-based active learning for multiclass brain-computer interface (BCI)

Brain-computer interface (BCI) algorithms based Support Vector Machine (SVM) and Naive Bayes (NB) give satisfactory performance but need a relatively large number of samples for training reliable classifier, which is difficult, expensive and time-consuming. In the paper, based on batch-mode active learning version, we propose the two algorithms, MS-based multiclass BCI algorithm and entropy-based multiclass BCI algorithm for solving multiclass BCI problems, which initially only need a small set of labeled samples to train classifiers. To assess the effectiveness of the two methods, we successfully test them with 9 subjects involved in the data set 2a of BCI Competition IV. The test results indicate that the exploitation of the two methods to unlabeled data can gradually improve classification results.