Research On Transductive Network Anomaly Detection Algorithm Based On Sample Scale Optimization

With the continuous development of the Internet and the popularity of computers,network security problems emerge one after another.As a dynamic network security defense technology,anomaly detection can make up for the defects of the traditional protection mechanism,and get more and more concerned by the experts in the field of network intrusion.Based on the research of Zheng Siwei and others,the main objective in this paper is to reduce false positive rate and computational overhead.The training sample set and the large network data feature dimension exists in TCM-RNE network anomaly detection method,resulting in the abnormal measurement is not accurate and the high rate of false positives,we put forward the corresponding optimization strategy and improved algorithm.The main research work is as follows:Based on particle swarm optimization principle,we proposed a sample scale optimization strategy and corresponding algorithm of particle swarm optimization for TCM-RNE.In view of the problem that the normal training set of TCM-RNE algorithm is too large,and the characteristics of fast convergence of particle swarm optimization,a small amount of high-quality samples are modeled from normal training samples.The comparison experiment on the UCI dataset shows that the particle swarm optimization strategy can effectively reduce the scale of training samples and the running time of the algorithm.We proposed a combined sample scale optimization strategy and corresponding algorithm based on ReliefF and CFS.Aim to dimension disaster occurred in entropy calculation of TCM-RNE algorithm,we combined the ReliefF algorithm and the CFS algorithm in filtering feature selection method,and the combined sample scale optimization algorithm is applied to the reduced sample feature space.The comparison experiment on UCI dataset shows that combinatorial sample scale optimization algorithm can effectively remove the sample independent features and redundant features at a small time cost.The TCM-RNE algorithm is optimized by particle swarm optimization strategy and combinatorial optimization strategy.We proposed a kind of TCM-RNE anomaly detection algorithm based on particle swarm sample scale optimization and a kind of TCM-RNE anomaly detection algorithm based on combinatorial sample scale optimization.At the same time,we compared the optimization strategies combined with TCM-KNN algorithm.The experiment on the KDD Cup1999 data set shows that the optimization strategy adopted in this paper also reduces the computational cost of the TCM-KNN and TCM-RNE algorithm training set,and the improved TCM-RNE algorithm has better detection performance.