Diabetes Diagnosis using Fuzzy Min-Max Neural Network with Rule Extraction and Apriori Algorithm

Abstract

This paper proposes a modified neural network called Fuzzy Min-Max Neural network (FMMN) that forms hyperboxes for classification and prediction and applied to Pima Indians Diabetes (PID) dataset. The modifications are made to improve its classification performance when a small number of large hyperboxes are formed in the network. This system is composed of formation of hyperbox, Pruning and Prediction and Rule Extraction. The hyperbox is formed by calculating its confidence factor. The user defined threshold is used to prune the hyper box with low confidence factors. The advantage of pruning is that it improves the FMMN performance during the large network of hyperbox formation also it facilitates the extraction of a compact rule set from FMMN to verify its prediction. Apriori algorithm is used to find closely related attributes using support and confidence factor. From closely related attributes a number of rules are generated. The proposed algorithm is applied on the PID dataset from standard University of California, Irvine (UCI) which demonstrates that the proposed system is useful for diabetes diagnosis and classification tool in real environments.