Distribution estimation based negative selection algorithm (DENSA)

The Negative Selection Algorithm is an anomaly detection technique, inspired by the self-nonself discrimination behavior observed in the Biological Immune Systems. The most controversial problem of the negative selection algorithm is its inherent limitation in detecting foreign patterns as anomalies. This limitation causes high false positive rate in anomaly detection systems which are based on the negative selection algorithm. To tackle this limitation, this paper introduces an efficient negative selection algorithm by focusing on generating more efficient detectors using a more flexible boundary for self-patterns. In other words, instead of applying conventional affinity measures, a Gaussian Mixture Model is fitted on normal space so that detectors are generated employing this Gaussian Mixture Model of self-space. The efficiency of the proposed algorithm is evaluated using different data sets including 2D synthesis data sets and NSL-KDD data set. The results indicate that generating detectors based on the Gaussian mixture model leads to more efficient negative selection algorithm even in real application with high dimensions where the traditional negative selection algorithms have limitations.