Abstract: A large quantity of holes is generated on the boundary of self and nonsclf region, because of the boundary dilemma of real-valued negative selection (RNS) algorithm. A real-valued negative selection algorithm with boundary delectors was presented. In the new approach,the detectors were interpreted using an aggressive interpretation,and during the training phase the boundary detectors whose aggressiveness was controlled by boundary threshold were generated and deployed on the boundary. The boundary detectors can reduce the holes on the boundary efficiently, and also probe the boundary between the self and nonsclf region implicitly. The algorithm was tested using both synthetic 2-D data set and MIT Darpa 1998 offline data set. Results demonstrate that the new approach has a much higher detection rate than RNS algorithm, in the case of the same false alarm rate, although it has a little higher minimum false alarm rate.

Key words: Artificial immune,Negative selection algorithm, Boundary dilemma, Hole