An Outlier Robust Negative Selection Algorithm Inspired by Immune Suppression

Abstract

The negative selection algorithm (NSA) is one of models in artificial immune systems. Traditional NSAs do not perform any differentiation for training self dataset and only use the mechanism of negative selection. They will generate excessive invalid detectors and have poor detection performance when the training selves contain noisy data. Inspired by immune suppression mechanism, an outlier robust NSA is proposed. The new algorithm will divide the training selves into internal selves, boundary selves and outlier selves. At the same time, the information hiding in different kind of selves is fully utilized. Furthermore, by combining negative selection mechanism with positive selection mechanism, the new algorithm can cover the non-self region more effectively. The experiment results show that no matter the training self data is clean or not, the new algorithm can obtain better detection performance with fewer detectors.