F1ow-based Anomaly Detection Using Multilayer Perceptron in Software Defined Networks
Yuan‐Cheng Lai, Kai-Zhong Zhou, Si-Ru Lin, Nai‐Wei Lo · 2019
For high-speed networks, this paper developed a flow-based anomaly detection system for reducing the overhead in Software Defined Networks (SDN). The controller in SDN uses a deep learning technique, Multilayer Perceptron (MLP), to automatically generate the weights for detecting the anomaly. We investigate the activation functions and the number of hidden layers used in MLP to compare flow-based MLP (FBM) and packet-based MLP (PBM). The results show that FBM is a better solution than PBM because it has lower false positive rate when true positive rate is high. Also FBM can provide lower overhead because PBM spends 123% time over FBM on establishing the MLP model.