Improving the efficiency of the artificial immune network algorithm (aiNet) in the radial basis function neural network construction

Abstract

The Artificial Immune Network (aiNet) is an Artificial Immune System algorithm for clustering tasks. Often, it requires significant processing time, as it is the case of applying in Radial Basis Function Neural Networks (RBFNN). In this work, to construct RBFNN applying the aiNet, attempting for minimize processing time, without increasing an error rate measure, three strategies are proposed: i) constructing RBFNN in each aiNet generation, preserving the RBFNN with the best performance; ii) refining the remained antibodies in each aiNet generation by the k-means algorithm; and iii) dividing the internal memory of clonal antibodies into smaller matrices, and submit them iteratively to the aiNet suppressing process. These strategies are showed suitable for minimizing the error rate and the processing time taking into account standard aiNet algorithm. In two RBFNN construction experiments, these strategies minimized the error rate and the processing time in 5,39% and 89,91%, respectively.