Projection with double nonlinear integrals for classification

Abstract

In this study, a new classification model based on projection with Double Nonlinear Integrals is proposed. There exist interactions among predictive attributes towards the decisive attribute. The contribution rate of each combination of predictive attributes, including each singleton, towards the decisive attribute can be re presented by a fuzzy measure. We use Double Nonlinear Integrals with respect to the signed fuzzy measure to project data to 2-Dimension space. Then classify the virtual value in the 2-D space projected by Nonlinear Integrals. In our experiments, we compare our classifier based on projection with Double Nonlinear Integrals with the classical method- Naïve Bayes. The results show that our classification model is better than Naïve Bayes. © 2008 Springer-Verlag.