Approximate Pattern Classification Using Neural Networks

In this paper, we propose an approximate classification method using neural networks for two-class classification problems. The con ventional classification problems can be described as finding a sharp (or crisp) boundary that divides a pattern space into two disjoint decision areas. On the other hand, our approximate classification problems can be viewed as finding a fuzzy boundary that divides the pattern space into three disjoint areas: two of them are the decision areas of the two classes and the other is the boundary area between them. The proposed method determines the three areas using two multilayer feedforward neural networks.