Intrusion Detection Based on Feature Selection and Temporal Convolutional Network in Mobile Edge Computing Environment

As an inevitable trend of future 5G networks, Mobile Edge Computing (MEC) has many advantages in providing content awareness and cross-layer optimization services. But it also faces malicious traffic attacks from emerging services and technologies. Intrusion Detection Systems (IDS) is an effective defense mechanism to monitor and detect abnormal data on the host side or network side. However, as the focus of research in network security, IDS can usually be deployed separately without collaboration. The previous work on IDS is mainly based on Recurrent Neural Network (RNN) design. However, there are characteristics of limited edge node resources in the MEC scenario, making it challenging to deploy IDS for MEC. Secondly, these RNN-based methods are not effective in terms of detection accuracy. We proposed the model with Extreme Gradient Boosting Decision Tree and Temporal Convolutional Network (XGBoost-TCN) to solve these problems. In short, we used the XGBoost algorithm to reduce high-dimensional traffic to low-dimensional traffic. After that, we used a TCN model to detect abnormal traffic. The effectiveness and adaptability of the model have been verified on the public dataset. In addition, the results of performance evaluation show that the model has higher detection accuracy than previous related work for highly imbalanced abnormal traffic datasets © 2022. International Journal of Network Security. All Rights Reserved.