An adaptive bandwidth scheduling algorithm for IoT devices using machine learning techniques
Kumar Saurabh, Satyasundara Mahapatra, Manish Tripathi · Franklin Open · 2026
The proliferation of the Internet of Things (IoT) has set off a hypergrowth in the number of connected devices, all competing for bandwidth in a network with limited capacity. Proper bandwidth deployment is crucial as it allows the system to function optimally with minimal latency and energy consumption. Conventional scheduling mechanisms often do not dynamically change the network in relation to the immediate conditions and, because of that, fail to operate properly. This paper is dedicated to the development of machine learning techniques in the field of adaptive bandwidth scheduling, which is the act of employing reinforcement learning methods to customize the distribution of bandwidth. The protocol is multifaceted: it adjusts to the priority of the devices, the traffic of data, and Quality of Service (QoS) demands, ensuring the use of the network to be at the optimum level in terms of minimal latency. To demonstrate, particularly, the learning strategy of automatic traffic prediction and the decision-making strategy of the RL agent that acts on the real-time system, the proposed algorithm proves to be an effective approach. In this context, the IoT devices are personified by agents, while the environment coincides with the network in which they are placed. The RL agent acquires the best bandwidth allocation scheme by communicating with the environment. Extensive simulations using benchmark IoT datasets demonstrate significant improvements in throughput, latency reduction, and bandwidth utilization efficiency compared to traditional static and heuristic-based scheduling algorithms. Apart from that, the proposed method and improvements showed that the efficiency could be increased by even 35% and latency could be reduced by 20% depending on the network's varying conditions. Based on the evaluation done on the testing, simulation that functions as the proposed algorithm's performance resulted differently in a simulation environment. The network has 100 devices of different types of traffic and bandwidth requirements. We, the authors, employed our proposed algorithm in our simulation, and further, we tried to align this choice to other traditional scheduling methods, namely Weighted Fair Queuing (WFQ) & First-Come-First Serve (FCFS), and a baseline Machine Learning (ML) algorithm. The results of the simulation experiments are indicative of the reduction of latency, increase of throughput, and the leading of energy efficiency compared to standard approaches. The proposed approach is at the top of the list in the area of IoT under consideration with its performances that are outstandingly better than the others; it is thus appropriate to be made use of in IoT problems that are more diverse