Multisource Neighborhood Immune Detector Adaptive Model for Anomaly Detection
Liang Xi,Rui-Dong Wang,Zhi-Yu Yao,Feng-Bin Zhang
DOI: https://doi.org/10.1109/tevc.2021.3058687
IF: 16.497
2021-06-01
IEEE Transactions on Evolutionary Computation
Abstract:The artificial immune system (AIS) is one of the important branches of artificial intelligence technology, and it is widely used in many fields. The detector set is the core knowledge set, and the AIS application effects are mainly determined by the generation, evolution, and detection of the detectors. Presently, the problem space (shape-space) of AIS mainly applied real-valued representation. But the real-valued detectors have some problems that have not been solved well, such as slow convergence speed of generation, holes in the nonself region, detector overlapping redundancy, dimension curse, etc., which lead to the unsatisfactory detection effects. Moreover, artificial immune anomaly detection is a dynamic adaptive model, needs to be evolved adaptively with the detection environments. Without better adaptive modeling, these problems mentioned before will get worse. In view of this, this article proposes a multisource immune detector adaptive model in neighborhood shape-space and applies it to anomaly detection: based on random, chaotic map and DNA genetic algorithm (DNA-GA), multisource neighborhood negative selection algorithm (MSNNSA), multisource neighborhood immune detector generation algorithm (MS-NIDGA), and neighborhood immune anomaly detection algorithm (NIADA) are proposed, so that the generation and detection of immune detectors can be improved efficiently; introducing immune adaptive and feedback mechanism, multisource neighborhood immune detector adaptive model (MS-NIDAM) is built, so that the detectors can be adaptively evolved in a more targeted search domain, and keep better distribution to the nonself region in real time, so as to solve various problems existing in the real-valued shape-space under dynamic environment mentioned before and improve the overall detection performances. The experimental results show that MS-NIDAM can improve the detector generation/evolution efficiency, keep the up-to-date understanding of the changing e-vironment, so as to obtain better overall detection performances and stability than other comparative methods.
computer science, artificial intelligence, theory & methods