SVDD-based Network Traffic Anomaly Detection Method with High Robustness
Shuqin Dong, Bin Zhang · 2019
Due to attacks' strong concealment and rarity, it is still a big challenge to discriminate anomaly in network traffic, especially in big data era where the amount of traffic and the feature dimensionality of each flow are high. In this paper, we proposed a network traffic anomaly detection method based on stacked denoising autoencoders (SDA) and support vector data description (SVDD). The method used SDA to extract deep traffic features with high robustness and reduce the dimensionality of each flow's original features, and trained SVDD with the deep features of normal network flows to construct a one-class classifier so that it can detect any network anomaly accurately. The experimental results using the NSLKDD dataset show that the proposed method has a higher detection performance and a lower time-consuming of training the classifier compared with single SVDD.