Research on the Multi-classifier Fusion Model Based on Choquet Integral

Abstract

In order to improve the performance of the classifiers, multi-classifier fusion is a good strategy. The multi-classifier fusion models based on Choquet integral have shown successful applications in practice since the adopted fuzzy measure in the integral can effectively describe not only the importance degree of the classifiers respectively but also the interaction among the classifiers. Considering many real problems, allowing fuzzy measure to assume interval or fuzzy numbers seems more reasonable. In this paper, we define a generalized Choquet integral in which an interval-valued fuzzy measure is adopted and its value is a crisp real number. Furthermore, the defined Choquet integral is applied in the multi-classifier fusion. PSO is used to determine the values of the interval-valued fuzzy measure. Experimental results demonstrate the rationality of the definition and the feasibility and effectiveness of the application in the multi-classifier fusion.KeywordsFuzzy measureChoquet integralInterval-valued Multi-classifier fusion