Research On Intrusion Detection Algorithm Based On BiRNN-SVM

People's Daily work and life are increasingly dependent on the network,and network security is more and more valued at the same time.Intrusion detection system can effectively detect and prevent abnormal behaviors and protect network users from the threat of malicious software.Currently,intrusion detection system has the problems of low detection rate and high false alarm rate.In order to solve the above problems,a Bi RNN-SVM algorithm was proposed for intrusion detection.Bi RNN generally adopts Soft Max classifier to fit the network traffic with timing.However,due to the same weight of all data participating in the training,it is easy to fall into the over-fitting state.SVM algorithm can combine the empirical risk and structural risk of support vector to avoid overfitting state,but the algorithm is worse than Bi RNN in feature extraction,so it is easy to fall into underfitting state.Therefore,Bi RNN-SVM algorithm is proposed to study network traffic data and identify various network behaviors.Among them,the single structure of Bi RNN adopts the classical LSTM structure in the RNN algorithm,the SVM algorithm is presented in the form of neural network,and the linear SVM is optimized by the random gradient descent method.Bi RNN-SVM algorithm is composed of forward LSTM and reverse LSTM,and the final layer is directly output without activation.Taking the folding loss of SVM algorithm as the optimization target,the algorithm is optimized by random gradient descent(SGD)optimizer.In terms of experimental data,cidds-001 data set will be used.In this data set,IP2 Vec technology will be adopted to characterize part of the text data,so as to compare the data processing scheme without IP2 Vec technology.In addition,two data sets containing normal and abnormal data and only abnormal data will be set.In terms of algorithm,16 experiments were designed based on Bi RNN-SVM algorithm,Bi RNN algorithm,SVM algorithm and Softmaxalgorithm.In terms of evaluation,the first is the observation generalization,the second is the accuracy,the recall rate and the f1 value,the third is the ROC curve,and the fourth is the detection rate.The experimental results show that birnn-svm algorithm has a strong generalization ability compared with other algorithms.Birnn-svm has high precision,recall rate and f1 value in various cases,and the accuracy rate is outstanding in all algorithms.Meanwhile,in the ROC curve,the AUC value is high,indicating that the algorithm has high robustness.