A compact fuzzy min max network with novel trimming strategy for pattern classification

Hyperbox classifier has large contribution to the field of pattern classification, because of its efficiency and transparency. Hyperbox classifier is efficiently implemented by using fuzzy min-max (FMM) neural network. FMM was modified many times to improve the classification accuracy. Moreover, there still exists a space for increasing the accuracy of hyperbox based classifiers. In this paper, four modifications are proposed to FMM network for increasing the classification accuracy rate. First, centroid and K-highest (CCK) based criteria to select the expandable hyperbox. Second, a new set of overlap test cases to consider all types of overlapping regions. Third, a new set of contraction rules to settle the overlapped regions. Fourth, novel hyperbox trimming strategy to reduce the system complexity. The proposed method is compared with FMM, enhanced FMM (EFMM) and Kn_FMM using five datasets. Experimental results clearly reflect the improved efficiency of proposed method. Proposed FMM (PFMM) network is also used to classify the histopathological images for knowing the best magnifying factor. (C)& nbsp;& nbsp;2022 Elsevier B.V. All rights reserved.

Automated summary

Hyperbox classifier has made significant contributions to the field of pattern classification due to its efficiency and transparency. This paper proposes four modifications to the fuzzy min-max (FMM) neural network for increasing the classification accuracy rate. Experimental results demonstrate the improved efficiency of the proposed method.