Artificial Immune Network Based Classification Algorithms And Applications

Artificial Immune Network (AIN) algorithm is the foundational algorithm which is based on Immune theory and especial the Idiotypic Immune Network (IIN) theory. And AIN has been applied into the field of intelligent computing. However, there are some shortcomings in the most existing AIN classification algorithms. On the one hand, the shortcomings lie in the mechanism of algorithm such as large-scale network, complex computation, and only once presenting the antigens, which resulted in the problem of low efficiency. On the other hand, they lie in the narrow applications, most AIN algorithms focus on the standard datasets classification and clustering.After analyzing above problems and aim to solve complex data classification and image classification problems, three AIN classification algorithms are proposed in this paper, the main work can be outlined as follows:(1) A new Artificial Immune Network classification algorithm is proposed. In the proposed algorithm, only one B-cell is used to denote single class in order to reduce the scale of network, and avoid the suppression operation between B-cells, moreover, a new affinity function based on the correct rate is proposed to realize the evaluation strategy based on antigen priority. The results of experiments indicate that the new algorithm has better accuracy and robustness.(2) A self-adaptive PSO based Artificial Immune Network classification algorithm is proposed. This method applies the self-adaptive PSO into the mutation process of the artificial immune network algorithm. Moreover, the strategy of every B-cell containing all types of information for antigens is used in the proposed algorithm. So the new algorithm has good global search ability and fast convergence speed.(3) An associate rules mining algorithm based on artificial immune network is proposed. This method introduces the association rules used in the data mining into the Artificial Immune Network algorithm. And it uses researching the optimal association rules to replace finding the best cluster center. Comparative experiments show that the method has satisfactory classification accuracy and convergence speed in dealing with multi-class and nominal attributes data.