Abstract

Network slicing (Ns) is a key enabling technology to support the concurrent provisioning of better quality of service (QoS) in 5G networks. These services have become essential for a telecom service provider (SP) to offer better QoS and QoE (quality of experience). The QoS parameters are used to estimate the performance of the network, and QoE determines user satisfaction with the network services. The main challenges faced by the service provider are to select the appropriate slice for each service and accurately classify these services on a timely basis to satisfy the Service level agreement (SLA) while improving the QoS and QoE. To overcome this issue, machine learning (ML) is a good solution. In this paper, we have proposed a 5G-KPQI (5G-key performance and quality indicator) model that considers the 5G service-based dataset for the 5G services classification. Next, we used feature selection (FS) methods to rank and select the best feature subset, which increases the performance of ML models and also reduces the training time required by the models. We subsequently considered various ML models to classify the services. Results demonstrate that the 5G-KPQI model ranks the features using Relief-F and mrMR methods and also reduces the training time of the model, hence improving classification performance measured by precision, accuracy, F1-score, recall, MCC, and time. The evaluation of the key approach outperforms in high classification accuracy and less training time using decision tree (DT) and random forest (RF).

中文翻译：

---

通过 5G 网络中的特征选择和机器学习对服务进行分类

摘要

网络切片 (Ns) 是支持在 5G 网络中同时提供更好的服务质量 (QoS) 的关键使能技术。这些服务对于电信服务提供商 (SP) 提供更好的 QoS 和 QoE（体验质量）至关重要。QoS参数用于评估网络的性能，QoE则确定用户对网络服务的满意度。服务提供商面临的主要挑战是为每项服务选择合适的切片，并及时准确地对这些服务进行分类，以满足服务级别协议（SLA），同时提高QoS和QoE。为了克服这个问题，机器学习（ML）是一个很好的解决方案。在本文中，我们提出了一种 5G-KPQI（5G 关键性能和质量指标）模型，该模型考虑基于 5G 服务的数据集进行 5G 服务分类。接下来，我们使用特征选择（FS）方法来排序并选择最佳特征子集，这提高了机器学习模型的性能，并减少了模型所需的训练时间。随后我们考虑了各种机器学习模型来对服务进行分类。结果表明，5G-KPQI 模型使用 Relief-F 和 mrMR 方法对特征进行排序，还减少了模型的训练时间，从而提高了通过精度、准确度、F1 分数、召回率、MCC 和时间衡量的分类性能。关键方法的评估结果优于使用决策树（DT）和随机森林（RF）的高分类精度和更少的训练时间。