A Comparative Study Of Machine Learning Strategies To Reduce Computational Impact And Improve Sustainable Network Management

Network management in the telecom industry is a resource-intensive task in which sub-optimal equipment configuration can significantly impact energy consumption. This research investigates machine learning enabled methods to optimise the allocation of CPU and memory dynamically, thereby minimising unused resources and overall energy expenditure. Using a simulated telecom dataset with real-time workload characteristics, we compare multiple machine learning models— Linear Regression, Random Forest, and XGBoost—to determine the most effective method for predictive resource allocation. Furthermore, we analyse the impact of data granularity by training models on 1-minute, 5-minute, and 15-minute interval datasets. Our findings indicate that XGBoost trained on 1-minute granularity data achieves the best performance, reducing CPU and memory over-provisioning while maintaining network performance. This research contributes to energy-efficient telecom network management, paving the way for future real-world AI-driven optimisations.