Misuse Intrusion Detection Based On Weighted Feature Selection

Along with the widely using of computer networks and the overrun of network attack behaviors, network security has become a focus and important part of national security. As an important technical measure to network security, intrusion detection is paid more attention by network users and researchers.In this thesis, a combination of classifiers and a data mining method are applied to network intrusion detection. Reference to the common intrusion detection framework (CIDF), a misuse intrusion detection model based on Random-Committee is proposed. The function, mechanism and realization of the model are discussed in this thesis.Knowledge used in this thesis found that connecting data in data mining and intrusion detection. These attack classes distribute unbalanced and the number of every classes are different largely in original training set. Therefore, the result of using these data are very bad. Hence, this thesis brings out a method to balance the training data. And this method can bring up the detected accurate of every class efficiently, shrink the size of training data, and make the system more efficiency by analyzing and matching.Feature ranking and selection are bought forward to filter the useless features of network connection. This will avoid the noise features, and also decrease the training time and space complexity.At last, this thesis advances this algorithm by feature selection and character sorting standard. Compare with methods before, it can get well test result.