A deep Q-learning framework for optimizing internet bandwidth management in higher education institutions in Uganda

Abstract

Internet bandwidth inefficiency remains a critical challenge in Ugandan Higher Education Institutions (HEIs), impairing academic delivery and administrative functions despite infrastructural investments. At Islamic University in Uganda (IUIU), static allocation policies result in severe congestion (user throughput collapsing to 20 Kbps during peaks) while 74% of provisioned bandwidth (306.7 Mbps) remains underutilised. This study designs a dynamic bandwidth allocation framework integrating organisational policy hierarchies with Deep Reinforcement Learning (DRL) to optimise resource distribution in real time. The framework classifies resources hierarchically, Academic (Teaching/Non-teaching) and Non-academic (Managerial/Non-managerial), with institutional priority weights (e.g., 70% bandwidth to Teaching during examinations). A Deep Q-learning agent dynamically adjusts allocation weights in response to simulated network conditions derived from IUIU’s historical traffic patterns. Trained over 50 epochs per academic season using a duelling network architecture with prioritised experience replay, the agent optimises a reward function balancing policy compliance and utilisation efficiency. Simulation results demonstrate 35% latency reduction during congestion events and 22% higher throughput for academic traffic compared to static baselines. The framework ensures critical services (e.g., learning management systems) receive guaranteed resources while improving overall bandwidth utilisation by 30%. Validated in IUIU’s operational context, this approach offers a scalable, policy-aware solution for resource-constrained HEIs, reducing dependency on costly infrastructure expansion. Contributions include a transferable methodology for embedding institutional priorities into technical resource governance; empirical validation of lightweight DRL in low-data environments; and a simulation toolkit calibrated to African HEI network dynamics. Keywords: Bandwidth optimisation, Deep Reinforcement Learning, Dynamic resource allocation, Higher Education Institutions, Quality of Service, Uganda