Analysis of Network Failure Detection Using Machine Learning in 5G Core Networks

Mobile service providers must consistently offer reliable and high-quality Internet services to support the 5G mobile network. Additionally, since the Internet is run cooperatively among the providers, an unexpected failure in a provider’s domain can rapidly spread all over the world, but only highly experienced operators can tackle these network failures. In order to address unexpected failures in the core network, machine learning plays an important role. Machine learning-based network operations operate efficiently and automatically, and it will also reduce operational costs. In this study, we used machine learning to analyze the network failures in the 5G core network. To identify the suitable approach, we analyze the performance of three ensemble learning-based machine learning algorithms—XGBoost, LGBM, and random forest, from different perspectives: preprocessing of training data, normal/abnormal samples, and feature importance. The results demonstrated that XGBoost provides higher accuracy with a smaller number of features. Similarly, LGBM and RF improve their performance while reducing the number of features. Overall, our proposed work has achieved a higher detection accuracy of 98.39% and a 100% detection rate in the three types of failures with XGBoost.