A real value negative selection algorithm based on antibody evolution for anomaly detection

Traditional negative selection algorithm generates immune detectors randomly, when the antigens distribute complicatedly and unevenly, the detectors are difficult to generate in the gap which between self samples and nonself samples. On the contrary, the detectors are redundant in the regions where...

Full description

Saved in:
Bibliographic Details
Published in2018 Tenth International Conference on Advanced Computational Intelligence (ICACI) pp. 692 - 699
Main Authors Yang, Tao, Chen, Wen, Liu, Zhengjun, Lin, Ping
Format Conference Proceeding
LanguageEnglish
Published IEEE 01.03.2018
Subjects
Anomaly detection
Cloning
Computer science
Detectors
evolution
Evolution (biology)
Immune system
negative selection
Training
Online AccessGet full text
DOI10.1109/ICACI.2018.8377545

Cover

More Information
Summary:Traditional negative selection algorithm generates immune detectors randomly, when the antigens distribute complicatedly and unevenly, the detectors are difficult to generate in the gap which between self samples and nonself samples. On the contrary, the detectors are redundant in the regions where antigens distribute sparsely. To deal with this limitation, this paper proposes a real value negative selection algorithm based on antibody evolution (AERNSA). AERNSA contains two kinds of antibody evolution: evolution based on "distribution of antigens" and evolution based on "antibody competition". The two kinds of evolution can guarantee the detectors to be effectively generated in the region where antigens distribute densely and to reduce the redundant detectors in the area where antigens distribute sparsely. The experimental results show that, on the artificial dataset Rectangle (2D) and the UCI standard datasets, compared with the classical RNSA and V-detector algorithm, AERNSA can significantly improves the detection rate with fewer detectors and shorter training times.
DOI:10.1109/ICACI.2018.8377545