With the popularity of Internet technology and network, network intrusion detection problem is highly concerned. Classification algorithm has been widely applied in network intrusion detection model and the algorithm based on the twin support vector machine becomes the hotspot. Least Squared Twin K-class Support Vector Classification(LST-KSVC) is popular for its running speed in solving the problem of multiple classification, but its empirical risk minimization principle results in a drawback in the capability of generalization prediction of new sample data. An improved least squares twin K-class support vector classification(ILST-KSVC) is proposed by using regularization method to punish the prediction coefficients of objective function, which will modify the deficiency of LST-KSVC, and its convergence is theoretically proved. To demonstrate the ability of classification and network intrusion detection of the improved algorithm, UCI data set and KDD CUP 99 data set are used in experiment. The results show that:(1)The ILST-KSVC proposed has a good effectiveness in classification;(2)Compared with the LST-KSVC model, ILST-KSVC has improved the accuracy by 1.41%;(3) In data preprocessing, employing the adaptive information entropy discretization method proposed will improve the accuracy of classification. |