UAV swarm communication reliability prediction using machine learning

Communication reliability is a key factor in enabling unmanned aerial vehicles (UAVs) to be deployed in many applications. Unlike conventional ground networks, UAV networks present unique challenges, experiencing diverse fading models and specific network traffic patterns. Many existing studies employing machine learning techniques often overlook the complete understanding of UAV communication reliability, which is defined not just by the probability of receiving packets correctly but also receiving them within a latency constraint. This research investigates employing a Bayesian network (BN) classification model and a logistic regression (LR) model as conditional probability estimators to not only predict UAV swarm communication reliability but also identify the most probable modes of communication failure. The models were trained utilizing a dataset generated through comprehensive network simulations reflecting different UAV network scenarios. The study shows that the trained BN model accurately predicts and classifies UAV swarm communication reliability, along with the most likely communication failure mode in various network scenarios. The trained BN model is useful for ensuring communication reliability during UAV mission planning and for maintaining the required level of communication reliability during UAV deployment.