Validation of the accuracy of clustering cardiac disease datasets with Cascaded ANFIS

Abstract

ANFIS is an AI that is easy to implement in hardware due to its operational structure. However, when the number of input items or classes to be classified increases, it becomes difficult to design and implement on hardware. Therefore, in this paper, we designed Cascaded ANFIS, which can be trained even with a large number of input items and classes, and trained a Heartdisease dataset with 13 input items, which is difficult to train with ordinary ANFIS, and compared its performance with that of other AIs. The results showed that the accuracy of the presence/absence of heart disease by Cascaded ANFIS can achieve good generalization performance compared to the results of other AIs. In addition, the results of examining the change in accuracy of Cascaded ANFIS when the number of input items was reduced showed that when diagnosing the level of heart disease, the six items selected according to certain criteria were more accurate, and the coefficient of determination, which is the criterion for evaluating the diagnostic result of level, was also higher.