Due to millions of heterogeneous physical nodes, multiple-vendor and multi-tenant domains, and technologies etc., 5G has greatly expanded the threat landscape. Particularly from the high rate of traffic and ultra-low latency requirement of applications in 5G networks, the detection of the network traffic anomalies in real-time is critical. The conventional security approaches lack compatibility with modern network designs and are not much effective in 5G settings. We propose a two-stage network traffic anomaly detection system compatible with ETSI-NFV standard 5G architecture. Our architecture consists of two modules, i.e., (a) Dimensionality Reduction to compress the sample size at the edge of 5G networks and (b) Deep Neural Network classifier (DNN) that detects traffic anomalies. We have conducted our experiments using OMNET++ and ETSI-NFV (OSM MANO) 5G orchestration real platform deployed on AWS cloud systems. We have used the UNSW-NB15 data set and have shown that at dimensionality reduction factor of 81% the detection accuracy obtained is 98%. The proposal is compared with other recent approaches to show the overall merit of the architecture.

中文翻译：

---

下一代网络降维入侵检测方案


由于数以百万计的异构物理节点、多供应商和多租户域以及技术等，5G极大地扩展了威胁范围。特别是5G网络中应用的高速率和超低时延要求，实时检测网络流量异常至关重要。传统的安全方法缺乏与现代网络设计的兼容性，并且在 5G 设置中效果不佳。我们提出了一种与 ETSI-NFV 标准 5G 架构兼容的两阶段网络流量异常检测系统。我们的架构由两个模块组成，即（a）用于压缩 5G 网络边缘样本大小的降维，以及（b）用于检测流量异常的深度神经网络分类器（DNN）。我们使用部署在 AWS 云系统上的 OMNET++ 和 ETSI-NFV (OSM MANO) 5G 编排真实平台进行了实验。我们使用了 UNSW-NB15 数据集，结果表明，在降维因子为 81% 时，获得的检测精度为 98%。该提案与其他最近的方法进行了比较，以显示该架构的整体优点。