Misclassification cost-sensitive fuzzy inference system based on support vector machines

Abstract

Under some restrictions, the functional equivalence between misclassification cost-sensitive support vector machines(MC-SVM) and rule-based fuzzy inference system(FIS) is proposed. Then based on the learning mechanism of MC-SVM, the algorithm of designing a rule-based FIS, misclassification cost-sensitive mercer binary FIS (MC-MBFIS), is given. The MC-MBFIS algorithm has the good generalization ability, can avoid the "curse of dimension", and has the transparent inference ability. Experimental results based on a few benchmark data sets show that the MC-MBFIS algorithm can reduce the average misclassification cost.