Mathematical programming and artificial neural network approaches to discriminant analysis

Discriminant analysis is widely used research tool in social sciences, in business areas such as finance, marketing, and accounting, and in other areas involving taxonomical and classification analyses such as biology. In this study, a new mathematical programming model and an artificial neural networks approach are proposed for classification problems. New mathematical programming model bases on the strong features of some mathematical programming models in the literature. Artificial neural networks have also a wide ranging usage area in the classification problems. Back-propagation algorithm, used in the training of the artificial neural networks, has negative features such as being captured in the local solutions and low performance of classification in some states. In this study, training of the artificial neural networks is implemented with real-coded genetic algorithm and network structure that has been trained with real-coded algorithm has been used in the solutions of the classification models. Both newly proposed mathematical programming model and network structure, and other conventional discriminant methods were tested by using 12 different real classification data taken from the literature and simulation data.The results show that classification success of new proposed mathematical programming model and artificial neural network approach trained with real-coded genetic algorithm are better than other classification methods.