Parameter analysis of negative selection algorithm

The performance of negative selection algorithm (NSA) is affected by several parameters such as its self radius and expected coverage. Traditionally, most of the parameters are selected based on experience, resulting in varying performances of NSAs. In the present study, the NSA parameters were analyzed based on a new set of evaluation criteria. The criteria were used to calculate the self radius by an iterative algorithm using some nonself samples as reference points for the nonself boundary. The difficult problem of estimating the overlap volume between immune hyperspheres was solved by the Monte Carlo method. It was found that the error of the estimated nonself coverage was the primary cause of the poor performance of some existing artificial immune systems. A confidence estimation method was thus developed for improving the estimation precision. Experiments were performed in which both fixed- and variable-radius detectors were generated using different parameter value combinations. The results revealed that a significantly higher NSA performance could be achieved by the proposed parameter calculation method. Statistical analysis of the experimental results further confirmed the effectiveness and practicability of the proposed method.