A Dynamic Recognition Neighborhood Based Immune Network Classification Algorithm and Its Performance Analysis

For lack of effective methods used by the traditional immune network algorithms to guide the memory cell determination, a dynamic recognition neighborhood based immune network classification algorithm is proposed. The algorithm uses a kernel function representation scheme to describe the antibody-antigen affinity, and constructs dynamic recognition neighborhood with using pair wise antigens to guide the antibody population evolution, in which the antibody nearest to the pairing antigen is determined as the memory cell. The algorithm is applied to multi-class problem and high dimensional classification problem to analyze the classification performance. Furthermore, the algorithm is used for many standard datasets classification to evaluate the algorithm overall performance. The results show that the proposed algorithm can achieve better classification performance, which indicates that the dynamic recognition neighborhood based training method is able to guide the memory cell generation effectively and improve the algorithm performance significantly.