A fast detector generation algorithm for negative selection

Abstract

Inspired by biological immune systems, the field of artificial immune system (AIS), particularly the negative selection algorithm (NSA), has been proved effective in solving computational problems. However in practical applications, NSA still encounter challenges, such as noise in training data leading to imprecise classifications, the lack of sufficient samples for detector maturation, and potential overlap among detectors. Address to these problems, we propose a novel hybrid detector generation algorithm based on fast clustering for artificial immune systems, namely FCAIS-HD. It primarily consists of two stages: first it utilizes a fast clustering algorithm to generate self-samples to decrease the effect of noise in the data. FCAIS-HD replaces the self-samples with a small number of self-detectors to reduce the time of generating non-self detectors; in the second stage, it utilizes a novel variable-radius non-self detector generation algorithm to generate a small number of non-self detectors with small overlap rates. Finally, both self-detectors as well as non-self detectors are used to implement hybrid detection (HD). Comprehensive experiments are conducted on both simulation and real world data sets to compare classification performance with baselines. The results demonstrate that FCAIS-HD outperforms other algorithms with well excluded low-level noise interference, higher rate of detection and less parameter sensitivity. Additionally, experiments are carried out on some real word data sets demonstrate that FCAIS-HD also performs well in high-dimensional data sets.