A fuzzy classifier based on partitioned hyperboxes
Ruck Thawonmas, Shigeo Abe · 2002
We discuss a method for improving the performance of a fuzzy classifier that approximates class regions in the input space directly using hyperboxes. The proposed method performs partition of the hyperboxes with the aim of preventing under-fitting of the training data due to large hyperboxes. It terminates according to a proposed terminating criterion that prevents over-fitting of the training data due to excessive partition. Experimental results on widely used iris data substantiate the effectiveness of the proposed method.