A Comparative Study of Machine Learning Models for Spreading Factor Selection in LoRa Networks

Low power wide area networks (LPWAN) technologies offer reasonably priced connectivity to a large number of low-power devices spread over great geographical ranges. Long range (LoRa) is a LPWAN technology that empowers energy-efficient communication. In LoRaWAN networks, collisions are strongly correlated with spreading factor (SF) assignment of end-nodes which affects network performance. In this work, SF assignment using machine learning models in simulation environment is presented. This work examines three approaches for the selection of the SF during LoRa transmissions: 1) random SF assignment, 2) adaptive data rate (ADR), and 3) SF selection through machine learning (ML). The main target is to study and determine the most efficient approach as well as to investigate the benefits of using ML techniques in the context of LoRa networks. In this research, a library that enables the communication between ML libraries and OMNeT++ simulator was created. The performance of the approaches is evaluated for different scenarios using the delivery ratio and energy consumption metrics.