A Dynamic Immune Algorithm with Immune Network for Data Clustering

This paper proposes a Dynamic Immune algorithm used for data clustering analysis. Its immune mechanism, partially inspired by self-organized mapping theory, is introduced to adjust the antibody's quantity and improve clustering quality. In order to guarantee clustering quality for highly non-linear distributed inputs, Kernel method is adopted to increase the clustering quality. In order to enhance direct descriptions about the clustering's center and result in input space, a new distance dimension instead of Euclidean distance is introduced by adopting Kernel substitution method while the training procedure is still running in input space. Simulation results are also provided to verify the algorithm's feasibility, clustering performance and anti-noise capability.