A negative selection algorithm with online adaptive learning under small samples for anomaly detection

The training stage and testing stage of traditional negative selection algorithm (NSA) are mutually independent, and NSA lacks continuous learning ability. Its detector cannot completely cover the non-self space. A new NSA with online adaptive learning under small training samples, OALI-detector, was proposed in this paper. I-detector can fully separate the self space from the non-self space with an appropriate self radius. It can adapt itself to real-time change of self space during the testing stage. The experimental comparison among I-detector, V-detector, and other anomaly detection algorithms in two artificial and Iris datasets shows that the I-detector can obtain the highest detection rate in most cases. The experimental comparison between OALI-detector and V-detector on Iris datasets shows that when overfitting does not occur, the OALI-detector can obtain the highest and lowest false alarm rates, even if only one self sample is used for training.