Classification based on Choquet integral

Abstract

The Choquet integral is a powerful aggregation tool in information fusing and data mining. In this paper, a generalized nonlinear classification model based on single Choquet integral is summarized, and a novel generalized nonlinear classification model based on cross-oriented Choquet integrals is presented. Compared to the classification model based on single Choquet integral, a couple of Choquet integral are used to achieve the classification boundaries which can classify data in such situation as one class surrounding another one in a high dimensional space. The classification problems come down to properly specifying the fuzzy measure on which the Choquet integral(s) are defined and the classifying boundaries by which the different classes are separated. The values of these unknown parameters are optimally determined by evolutionary computation. The performance of these models are compared and validated with existed methods on a number of benchmark datasets.