Multiple Classifiers Combination Based on Interval-valued Fuzzy Permutation

Multiple classifiers combination is a technique that combines the decisions of different classifiers as to reduce the variance of estimation errors and improve the overall classification accuracy. A new multiple classifiers fusion method integrated classifier selection and classifier combination is proposed in this paper. It is base on interval-valued fuzzy permutation. Firstly, normalize all classifier posterior probabilities using the priori knowledge of corresponding classifier recognition rate. And then, convert the decision matrix of multiple classifier system into interval-valued fuzzy decision matrix. Thirdly, determine the grade of possibility of each class for input sample in multiple classifier system. Finally, select the best classifier in current pattern recognition task using interval-valued fuzzy permutation and use the best classifier to make final decision. The experiments have shown that the new multiple classifiers fusion approach using interval-valued fuzzy permutation can provide much better accuracy compared to independent classifiers and some other fusion methods.