A Batch-mode Active Learning Method Based on the Nearest Average-class Distance (NACD) for Multiclass Brain-Computer Interfaces

In this paper, a novel batch-mode active learning method based on the nearest average-class distance (ALNACD) is proposed to solve multi-class problems with Linear Discriminate Analysis (LDA) classifiers. Using the Nearest Average-class Distance (NACD) query function, the ALNACD algorithm selects a batch of most uncertain samples from unlabeled data to improve gradually pre-trained classifiers' performance. As our method only needs a small set of labeled samples to train initial classifiers, it is very useful in applications like Brain-computer Interface (BCI) design. To verify the e®ectiveness of the proposed ALNACD method, we test the ALNACD algorithm on the Dataset 2a of BCI Competition IV. The test results show that the ALNACD algorithm o®ers similar classification results using less sample labeling e®ort than Random Sampling (RS) method. It also provides competitive results compared with active Support Vector Machine (active SVM), but uses less time than the active SVM in terms of the training.