An approach of constructing ECOC adaptively based on confusion matrix

Error correcting output codes(ECOC) is an effective frame that can decompose a multiclass problem into a set of complementary two-class problems, and the research of encoding based on data especially attracts attentions. In this paper, we propose a new encoding method CMECOC for constructing ECOC adaptively based on confusion matrix, first, we obtain the separability measure between each pair of patterns with the help of confusion matrix, then abide by Fisher´s rule, the most favorable combination of patterns for classification can be found, at last, we get binary partitions based on the way of pattern combination, and a data driven coding matrix can be achieved. Experimental results on UCI datasets with support vector machine(SVM) as the binary classifier show that our approach can provide a better performance and robustness of classification with a little longer but acceptable coding length.