A negative selection algorithm based on hierarchical clustering of self set

Abstract

Negative selection algorithm (NSA) is an important method of generating artificial immune detectors. However, the traditional NSAs aim at eliminating the self-recognized invalid detectors, by matching candidate detectors with the whole self set. The matching process results in extremely low generation efficiency and significantly limits the application of NSAs. In this paper, an improved NSA called CB-RNSA, which is based on the hierarchical clustering of self set, is proposed. In CB-RNSA, the self data is first preprocessed by hierarchical clustering, and then replaced by the self cluster centers to match with candidate detectors in order to reduce the distance calculation cost. During the detector generation process, the candidate detectors are restricted to the lower coverage space to reduce the detector redundancy. In the paper, probabilistic analysis is performed on non-self coverage of detectors. Accordingly, termination condition of the detector generation procedure in CB-RNSA is given. It is more reasonable than that of traditional NSAs, which are based on predefined detector numbers. The theoretical analysis shows the time complexity of CB-RNSA is irrelevant to the self set size. Therefore, the difficult problem, in which the detector training cost is exponentially related to the size of self set in traditional NSAs, is resolved, and the efficiency of the detector generation under a big self set is also improved. The experimental results show that: under the same data set and expected coverage, the detection rate of CB-RNSA is higher than that of the classic RNSA and V-detector algorithms by 12.3% and 7.4% respectively. Moreover, the false alarm rate is lower by 8.5% and 4.9% respectively, and the time cost of CB-RNSA is lower by 67.6% and 75.7% respectively.