Machine learning aided scheme for load balancing in dense IoT networks

Abstract

With the dramatic increase of connected devices, the Internet of things (IoT) paradigm has become an important solution in supporting dense scenarios such as smart cities. The concept of heterogeneous networks (HetNets) has emerged as a viable solution to improving the capacity of cellular networks in such scenarios. However, achieving optimal load balancing is not trivial due to the complexity and dynamics in HetNets. For this reason, we propose a load balancing scheme based on machine learning techniques that uses both unsupervised and supervised methods, as well as a Markov Decision Process (MDP). As a use case, we apply our scheme to enhance the capabilities of an urban IoT network operating under the LoRaWAN standard. The simulation results show that the packet delivery ratio (PDR) is increased when our scheme is utilized in an unbalanced network and, consequently, the energy cost of data delivery is reduced. Furthermore, we demonstrate that better outcomes are attained when some techniques are combined, achieving a PDR improvement of up to about 50% and reducing the energy cost by nearly 20% in a multicell scenario with 5000 devices requesting downlink traffic.