Multi Class Multi Label Based Fuzzy Associative Classifier with Genetic Rule Selection for Coronary Heart Disease Risk Level Prediction

Abstract

Most of the associative classifiers are crisp in nature. They focus to develop two class classifier. Fuzzy based approach in associative classifier improves the accuracy of the classifier and also helps in developing classifier with more than two classes. According to the World Health Organization (WHO), coronary heart disease accounts for about 17 million (approximately 30%) deaths annually throughout the world. Previous works in the heart disease risk level prediction concentrated only two classes whether a person has a possibility of heart disease or not. In the field of medical, early detection of diseases helps in curative of the diseases. Hence, developing a classifier by considering more than two classes improves the medical decision support system. Our proposed clinical decision support system helps in identifying coronary heart disease at the early stage also. In this paper, we propose a classifier framework for various risk level prediction, which consists of three major components. The first component is fuzzification of a data set by using various membership function based on the attribute nature. Second is generating fuzzy rules with multi class multi label based associative classifier. Associative classifier aims to discover a small set of rules that forms an accurate classifier. We extended multi class multi label associative classifier to build classifier for more than two risk levels. Final component is a hybrid model for rule selection which is a combination of pre selecting the best rules with best confidence value and optimizing rule selection using genetic algorithm. Experimental results show the effectiveness of our proposed system