RESEACH OF INTRUSION DETECTION BASED ON COST-SENSITIVE

Most traditional intrusion detection systems attempt to achieve a low error rate. Actually, different mistakes will lead to different losses. The standard classifier based on minimizing the error rate is quite unreasonable and unable to obtain a good classification and prediction results. Cost-sensitive mechanism by introducing proposed to minimize the overall cost of the misclassification. With K-nearest neighbors and Bayesian decision theory, the algorithm is improved to avoid the complicated calculation process. The experimental result demonstrate that we can trade off between false negative and false positive by cost matrix, which can minimize false negative and while constraining false positive at a low level so as to minimize the total misclassification cost.