Networks, threat models, and malicious actors are advancing quickly. With the increased deployment of the 5G networks, the security issues of the attached 5G physical devices have also increased. Therefore, artificial intelligence based autonomous end-to-end security design is needed that can deal with incoming threats by detecting network traffic anomalies. To address this requirement, in this research, we used a recently published 5G traffic dataset, 5G-NIDD, to detect network traffic anomalies using machine and deep learning approaches. First, we analyzed the dataset using three visualization techniques: t-Distributed Stochastic Neighbor Embedding (t-SNE), Uniform Manifold Approximation and Projection (UMAP), and Principal Component Analysis (PCA). Second, we reduced the data dimensionality using mutual information and PCA techniques. Third, we solve the class imbalance issue by inserting synthetic records of minority classes. Last, we performed classification using six different classifiers and presented the evaluation metrics. We received the best results when K-Nearest Neighbors classifier was used: accuracy (97.2%), detection rate (96.7%), and false positive rate (2.2%).

翻译：网络、威胁模型及恶意攻击者正在快速演进。随着5G网络部署规模的扩大，其关联物理设备的安全问题亦随之增加。因此，亟需构建基于人工智能的自主端到端安全体系，通过检测网络流量异常以应对潜在威胁。为满足此需求，本研究采用最新发布的5G网络流量数据集5G-NIDD，运用机器学习与深度学习方法检测网络流量异常。首先，我们通过三种可视化技术进行分析：t分布随机邻域嵌入（t-SNE）、均匀流形逼近与投影（UMAP）及主成分分析（PCA）。其次，利用互信息与PCA技术降低数据维度。再次，通过插入少数类合成记录解决类别不平衡问题。最后，采用六种不同分类器进行分类并呈现评估指标。使用K近邻分类器时取得最佳结果：准确率（97.2%）、检测率（96.7%）及假阳性率（2.2%）。