Machine-Learning-Based Predictive Handover

Good mobility performance is critical in cellular networks for ensuring that each user is always connected to the best possible cell and that the handovers are executed timely to minimize radio link failures while avoiding unnecessary handovers. A potential new approach to this challenge is to learn the local radio conditions and adapt and customize the mobility to them. This work proposes and investigates a machine learning method for learning the optimal time and destination for handovers in 5G radio networks, as well as how to use the learned model to trigger handovers based on the predicted radio conditions. The complete solution is analyzed and compared to the state of the art mobility methods to evaluate its performance in reducing the system total outage.