Improving the classification accuracy of cardiac patients

In medicine, the cost of erroneously classifying an ill subject as healthy could have disastrous consequences. Many classification techniques however, do not try to improve the classification on one of the classes involved. In this paper different strategies are proposed to solve this problem using artificial neural networks (NN), and they are applied to a heart disease dataset. An important reason for using NN is that the methods the authors propose can be directly incorporated into the learning process. To achieve this, two forms of NN training are used. One involves a slight modification of the usual backpropagation algorithm (BP), and in the other, the training is achieved through an evolution program (EP). The latter is applied to take advantage of their ability to handle more complicated fitness functions.