GraySOFM network for solving classification problems

In this paper, we incorporate gray relational pattern analysis into the self-organizing feature maps (SOFM) network to develop a GraySOFM network. A hybrid neighborhood function is proposed to select the nodes that need to be updated. Only high-related neighboring nodes around the winning node are updated to reduce the computational load and to achieve a better convergence of the training process. In the learning rule, we also take the pattern relationships into account. The GraySOFM is applied to solve one classification problem and five traveling salesman problems as examples. Computer simulations show better results than that of other known SOFM networks.