An efficient negative selection algorithm with further training for anomaly detection

Negative selection algorithm has been shown to be efficient for anomaly detection problems. This letter presents an improved negative selection algorithm by integrating a novel further training strategy into the training stage. The main process of further training is generating self-detectors to cover the self-region. A primary purpose of adopting further training is reducing self-samples to reduce computational cost in testing stage. It can also improve the self-region coverage. The testing stage focuses on the processing of testing samples lied within the holes. The experimental comparison among the proposed algorithm, the self-detector classification, and the V-detector on seven artificial and real-world data sets shows that the proposed algorithm can get the highest detection rate and the lowest false alarm rate in most cases.