A self region based real-valued negative selection algorithm

Point-wise negative selection algorithms, which generate their detector sets based on point of self da-ta, have lower training efficiency and detection rate. To solve this problem, a self region based real-valued neg-ative selection algorithm is presented. In this new approach, the continuous self region is defined by the collec-tion of self data, the partial training takes place at the training stage according to both the radius of self region and the cosine distance between gravity of the self region and detector candidate, and variable detectors in the self region are deployed. The algorithm is tested using the triangle shape of self region in the 2-D complement space and KDD CUP 1999 data set. Results show that, more information can be provided when the training self points are used together as a whole, and compared with the point-wise negative selection algorithm, the new ap-proach can improve the training efficiency of system and the detection rate significantly.