A self-adaptive negative selection algorithm used for anomaly detection

Abstract

A novel negative selection algorithm (NSA), which is referred to as ANSA, is presented. In many actual anomaly detection systems, the training data are just partially composed of the normal elements, and the self/nonself space often varies over time. Therefore, anomaly detection system has to build the profile of the system based on a part of self elements and adjust itself to adapt those variables. However, previous NSAs need a large number of self elements to build the profile of the system, and lack adaptability. In order to overcome these limitations, the proposed approach uses a novel technique to adjust the self radius and evolve the nonself-covering detectors to build an appropriate profile of the system. To determine the performance of the approach, the experiments with the well-known dataset were performed. Results exhibited that our proposed approach outperforms the previous techniques.