5G cellular networks come with many new features compared to the legacy cellular networks, such as network data analytics function (NWDAF), which enables the network operators to either implement their own machine learning (ML) based data analytics methodologies or integrate third-party solutions to their networks. In this paper, the structure and the protocols of NWDAF that are defined in the 3rd Generation Partnership Project (3GPP) standard documents are first described. Then, cell-based synthetic data set for 5G networks based on the fields defined by the 3GPP specifications is generated. Further, some anomalies are added to this data set (e.g., suddenly increasing traffic in a particular cell), and then these anomalies within each cell, subscriber category, and user equipment are classified. Afterward, three ML models, namely, linear regression, long-short term memory, and recursive neural networks are implemented to study behaviour information estimation (e.g., anomalies in the network traffic) and network load prediction capabilities of NWDAF. For the prediction of network load, three different models are used to minimize the mean absolute error, which is calculated by subtracting the actual generated data from the model prediction value. For the classification of anomalies, two ML models are used to increase the area under the receiver operating characteristics curve, namely, logistic regression and extreme gradient boosting. According to the simulation results, neural network algorithms outperform linear regression in network load prediction, whereas the tree-based gradient boosting algorithm outperforms logistic regression in anomaly detection. These estimations are expected to increase the performance of the 5G network through NWDAF.

中文翻译：

---

使用机器学习的 5G 蜂窝网络中的智能网络数据分析功能

与传统蜂窝网络相比，5G 蜂窝网络具有许多新功能，例如网络数据分析功能 (NWDAF)，它使网络运营商能够实施自己的基于机器学习 (ML) 的数据分析方法或集成第三方解决方案到他们的网络。本文首先描述了第三代合作伙伴计划（3GPP）标准文档中定义的NWDAF的结构和协议。然后，根据 3GPP 规范定义的字段生成 5G 网络的基于小区的合成数据集。此外，一些异常被添加到该数据集中（例如，特定小区中的流量突然增加），然后对每个小区、订户类别和用户设备内的这些异常进行分类。之后，三个 ML 模型，即 实施线性回归、长短期记忆和递归神经网络来研究 NWDAF 的行为信息估计（例如，网络流量异常）和网络负载预测能力。对于网络负载的预测，使用三种不同的模型来最小化平均绝对误差，这是通过从模型预测值中减去实际生成的数据来计算的。对于异常的分类，使用了两个 ML 模型来增加接收器操作特征曲线下的面积，即逻辑回归和极端梯度提升。根据仿真结果，神经网络算法在网络负载预测方面优于线性回归，而基于树的梯度提升算法在异常检测方面优于逻辑回归。