Research On Network Traffic Anomaly Detection Method Based On Multiple Mixed Features And LSTM

With the continuous update and upgrading of digital technology,the Internet is closely related to people’s study and work.At the same time,a secure and orderly network environment is extremely important.In recent years,network attacks emerge in an endless stream,posing a great threat to Internet security.Due to the continuous development and application of deep learning in various fields,its research in the field of network security is also increasing.Therefore,in view of the attack events in the Internet,this paper takes network traffic as the research object.Firstly,the characteristics of network traffic are selected.Secondly,the neural network in deep learning is used as the anomaly detection engine,so as to realize the discrimination and detection of abnormal traffic.Firstly,considering that network traffic has many redundant feature representations,this paper proposes a multiplex hybrid feature selection algorithm to filter the original feature set of network traffic,so as to obtain a simplified feature subset.The hybrid feature selection algorithm uses three fusion algorithms to analyze traffic from two dimensions.Among them,extreme random tree and limit gradient lifting tree algorithms can calculate and compare the importance score of each feature and analyze traffic from the dimension of feature and feature.Pearson algorithm calculates the correlation coefficient between them from the dimension of feature and category.The three fusion algorithms can comprehensively evaluate the features,and a new feature importance ranking can be obtained by combining the feature importance and correlation coefficient,according to which feature selection can be carried out.Secondly,combined with the multi-hybrid feature selection algorithm proposed above,this paper proposes a Long Short-Term Memory network(LSTM)anomaly detection method with optimized parameters.In this method,LSTM neural network is used as anomaly detection engine,and an improved Grasshopper Optimization Algorithm(GOA)is designed to optimize the superparameters of LSTM neural network.Thus,a LSTM anomaly detection engine based on improved GOA parameter optimization is implemented,which can effectively detect and discriminate network traffic.Finally,by setting up the experimental environment,the performance of the proposed feature selection algorithm and network traffic anomaly detection engine is verified and analyzed.Experiments show that compared with other anomaly detection methods and different parameter optimization methods,the anomaly detection method proposed in this paper has higher detection indexes.