Abstract: Intrusion detection system based on core vector machine with extensive kernel methods is presented to get rid of the restriction of kernels and the sub-quadratic problem. Firstly, the center-constrained minimum enclosing ball of a training data set is solved by the algorithm. The new MEB (minimum enclosing ball) is obtained by the simple update of the center and radius of the ball. Then the optimal separating hyperplane is constructed by the solutions of the core set. The convergence, time complexity and space complexity are proved theoretically. Finally, according to the distribution of the core set, the different intrusion actions can be detected. The related experiment indicates that the algorithm is feasible and effective.