Hybrid Artificial Immune Algorithm and CMAC Neural Network Classifier for Supporting Business and Medical Decision Making

Abstract

Decision making that involves credit scoring and medical diagnosis can be considered to solve classification problems. Among the many data miming (DM) methods that have been developed to solve these classification problems are neural network (NN) and support vector machine (SVM) classifiers. Despite their successful application to classification problems, these classifiers are limited, in that users must use trial-and error to modify specific parameter settings. Fortunately, the setting of the parameters for those classifiers can be viewed as an unconstrained global optimization problem. To overcome this limitation of those classifiers, this work develops an advanced DM method that combines an artificial immune algorithm (AIA) and a MIMO cerebellar model articulation controller NN (CMAC NN) classifier (AIA-MIMO CMAC NN classifier). The AIA is a stochastic global optimization method and its parameters are easily set. The proposed CMAC NN classifier is characterized by its fast learning, reasonable generalization ability and robust noise resistance. The proposed AIA-MIMO CMAC NN classifier uses an outer AIA to optimize the parameter settings of an inner MIMO CMAC NN classifier, which is used to solve classification problems. The performance of the proposed classifier is also evaluated using a set of real-world classification problems, such as credit scoring and medical diagnosis. Moreover, this work compares the numerical results obtained using the proposed AIA-MIMO CMAC NN classifier with those obtained using published classifiers (such as SVM, SVM-based classifiers, NN classifiers and C4.5). Experimental results indicate that the classification accuracy of the proposed AIA-MIMO CMAC NN classifier is superior to those of some published classifiers. Hence, the AIA-MIMO CMAC NN classifier can be viewed an alternative DM method for supporting business and medical decision making.