A Novel Compact Rule Extractor Based on Genetic-Fuzzy Algorithm

Abstract

Extracting effective rules in medical data with two indicators of accuracy and high interpretability is essential in increasing the accuracy and speed of diagnosis by specialists. As a result, the production of decision support systems that are able to detect data-driven rules play a vital role in the early detection of disease, even in areas where there is no access to a specialist. In this paper, a novel automatic rule extractor is presented using a hybrid model consisting of fuzzy logic and evolutionary algorithm. Fuzzy systems are suitable for making diagnostic models due to the high interpretability of their rules. The genetic algorithm is used to automatically generate these rules. To evaluate the proposed method, Pima Diabetes dataset including 768 records and 9 variables was used. The accuracy of the proposed model on the PIMA dataset was 77.12%. This is achieved by 7 fuzzy rules with an average length of 2.1, using three linguistic variables that represent low, normal and high values of each of the independent variables. All membership functions are the same width. According to the three criteria of low number of rules, short rule length and symmetric membership functions with the same width, the proposed method is quite suitable for extraction of compact rule base with high accuracy and interpretability in medical data.