Dynamic proactive–reactive scheduling for URLLC in 5G: Leveraging XGBoost and network virtualization

In this paper, we focus on service coexistence within the 5G network. Specifically, we aim to enhance the Quality of Service (QoS) for Ultra-Reliable Low Latency Communications (URLLC) within 5G networks. We propose a smart scheduler by integrating machine learning models. Real-time channel reports are used to make data-driven decisions based on current network conditions, thereby centralizing scheduling tasks to streamline operations and enhance adaptability.
The key contributions include the development of an adaptive scheduling strategy that dynamically switches between reactive and proactive approaches, ensuring optimal balance between low latency and ultra-reliability for URLLC services. Additionally, the Extreme Gradient Boosting (XGBoost) algorithm is applied to accurately predict URLLC traffic delays, assuring effective prioritization and timely delivery of time-sensitive services. Furthermore, the proposed solution combines network virtualization and machine learning techniques to address security risks and potential service interruptions linked to reactive scheduling policies. The proposed approach enhances existing methods by improving network virtualization features, which are integrated with machine learning models, specifically XGBoost to optimize the scheduling of ultra-reliable low-latency communication (URLLC). Unlike previous work, our proposal uses network virtualization to enable a dynamic combination of proactive and reactive scheduling strategies that can be adapted in real-time. Network virtualization features provide centralized control and resource management, resulting in reduced complexity and more efficient resource allocation. This helps the scheduler react quickly to changing network conditions, ensuring low latency and high reliability of URLLC services. In addition, network virtualization enhances the security level by isolating different types of traffic and mitigating risks such as Denial of Service (DoS) and impersonation attacks. Subsequently, the integration of advanced network virtualization functions with machine-learning predictions not only optimizes planning but also enhances security and resource management.
The proposed system is evaluated through simulations, demonstrating robust performance with low latency and high reliability for URLLC services. The XGBoost model achieves impressive results with low Root Mean Squared Error (RMSE) and Mean Absolute Error (MAE) on both validation and holdout datasets.