Hybrid Real-coded Genetic Algorithm and MIMO CMAC NN Classifier for Solving Medical Data Classification Problems

Abstract

Medical diagnosis is widely viewed as binary classification problems. To reduce classification error and parametrization of an efficient classifier, this work develops a hybrid real-coded genetic algorithm (RGA) and MIMO cerebellar model articulation controller neural network (CMAC NN) classifier. The parameter settings of the MIMO CMAC NN classifier are optimized using the RGA approach. Classification problems are then solved using the MIMO CMAC NN classifier. The performance of the proposed RGA-MIMO CMAC NN classifier is evaluated using two real-world datasets, i.e. diabetes and cancer datasets. The classification errors obtained using the RGA-MIMO CMAC NN classifier are compared with those obtained using individual MIMO CMAC NN classifier and published classifiers (e.g., genetic programming-based, NN-based and GA-based classifiers) for diabetes and cancer datasets. Experimental results indicate that the classification errors obtained using the proposed RGA-MIMO CMAC NN classifier are smaller than those of some individual and hybrid published classifiers. Moreover, the proposed approach can reduce parametrization of the MIMO CMAC NN classifier. Hence, the proposed RGA-MIMO CMAC NN classifier is highly promising for use as alternative classifier for solving medical data classification problems.