Research on Fusion Neural Network Intrusion Detection Based on Improved WGAN Algorithm
Aiming at the problem that existing intrusion detection models pay more attention to the overall detection rate of samples and ignore the categories of few-sample attacks,an improved Wasserstein Generative Adversarial Network(WGAN)algorithm has been proposed,which can expand the categories of few-sample attacks to balance the normal samples and attack samples,reducing the interference caused by class imbalance to the model.Firstly,the model collapse of the original WGAN caused by improper selection of weight clipping parameters is avoided,by introducing regularization terms on the discriminator.At the same time,in order to improve the generation quality,a new information loss function is added to the generator and cosine similarity is used to measure the deviation between real data and generated data.Then,a fused neural network model is used to learn high-dimensional features of the balanced dataset.Finally,the method is applied on the standard data set UNSW-NB15 to verify its performance.The results show that compared with traditional WGAN,the precision of the improved WGAN increases by 3.01%,AUC increases by 4.00%,and accuracy increases by 5.46%;compared with a single classification model,the F1 scores on the few-sample categories Analysis,Backdoor,Worms,and Shellcode increases by 8.36%～13.20%,11.48%～13.28%,54.16%～58.63%,and 28.67%～42.85%,respectively.
万方数据
