From Prediction to Protection: A Network Digital Twin framework for Traffic Monitoring and Anomaly Detection

The rapid growth of multimedia applications, cloud services and machine to machine communications has created dynamic and unpredictable traffic patterns in mobile networks. Understanding these patterns and reacting quickly to anomalies are essential for maintaining service quality and security. Digital Twins (DTs) are emerging as an enabler technology for network emulation, offering a virtual environment for what-if case scenario planning, impact analysis and closed-loop control. In this paper, we propose a Network Digital Twin (NDT) framework that simultaneously forecasts key network traffic operational indicators, the ratio of TCP to UDP packets and the ratio of TCP to UDP bytes and further we use the forecasting errors to detect anomalies in near real-time. Our proposed twin integrates deep learning and ensemble models, provides an interactive interface for configuring observation windows, training cycles and prediction horizons, and supporting anomaly injection to test detection performance. We further enhance the framework by integrating inference with new data to evaluate excellent performance. Experimental findings show excellent prediction with errors RMSE and MAE values as low as 0.09-0.10 and anomaly detection performance above 80 %. These results demonstrate how network digital twins can move from prediction to protection, enabling proactive network resilience.