Building an Application to Monitor Abnormal Network Node Traffic using Graph Learning Method

Abstract

The authors have developed an advanced monitoring application based on state-of-the-art graph learning methods to enhance the management and operational safety of telecommunications and network systems in Vietnam. This system integrates Long Short-Term Memory (LSTM) [6], Graph Convolutional Networks (GCNs) [5], and Graph Attention Networks (GATs) [4] to model and analyse the dynamic behaviour of network nodes. By combining temporal and structural features of network data, the application is capable of detecting anomalies, forecasting potential failures, and recommending timely interventions with minimal human oversight. These predictive capabilities significantly enhance the reliability, efficiency, and safety of telecommunications infrastructure, while also reducing downtime and lowering operational and maintenance costs. Additionally, the authors have developed custom software tools to support the labelling of raw datasets collected from network nodes. These datasets originate from existing static alerting systems used by network operators. The labelling tool enables efficient and consistent annotation of data, which is crucial for training and validating machine learning models. By transforming unstructured raw logs into structured labelled data, the system ensures higher accuracy in learning algorithms. Overall, this solution offers a practical and intelligent approach to managing large-scale network systems, particularly in developing regions like Vietnam, where automation and cost efficiency are critical.