Towards zero downtime: Using machine learning to predict network failure in 5G and beyond

A stable network is very important to both network providers and their customers, as it increases reliability, improves security and helps customers and companies save costs. When network outages occur, they result in significant downtime and financial losses for organizations and network users. Traditional methods of detecting and troubleshooting network failures are often reactive and time-consuming, whereby network administrators rely on traditional methods such as reactive monitoring and manual troubleshooting. These methods are often not effective in detecting and preventing network failures. In this paper, we propose a machine learning-based approach to predict network failures and minimize downtime. Network performance observability data from a 5G core network testbed based on Cloud-native Network Functions (CNFs) is used to train several supervised learning models, including random forest, gradient boosting regressor, conventional support vector regressor and proposed support vector regressor, to predict network failures. Our experiments and analysis show that the proposed model Support Vector Regressor (SVR) produced better results as compared to other models. In a very short amount of time (ten seconds), the proposed SVR model is capable of predicting whether a network failure event will occur or not within the next ten minutes, with an f1-score of more than 0.9. Our results indicate that machine learning-based approaches can significantly enhance the detection and prediction of network failures, leading to zero downtime and improved network performance.