A boundary based classifier combination method
Ming Liu, Kunlun Li, Rui Zhao · 2009
In this paper, a new classifier combination method is proposed for two-class problems. The boundaries of the classes are extracted directly from the given training set, and a set of linear combination rules are defined based on each sample on the class boundaries. The new approach is tested on two large public datasets, and the experimental results show its good performances. Comparing with combination methods such as linear combination, voting, decision templates, our method has higher classification accuracy; comparing with the k-NN rule, its computational complexity is much lower.