Machine Learning-Based Communication Collision Prediction and Avoidance for Mobile Networks

Abstract

Communication collision between devices represents a critical issue in wireless networks. It can cause network disruption, packet loss, communication delay and energy wastage. In the case of mobile networks, collision occurs frequently because of nodes’ mobility. Existing works rely on resolving collision after it happens or use GPS coordinates to predict neighborhood status in the future, and both methods cause collision. Those solutions cause packet loss and/or energy wastage.

In this paper, we introduce a GPS-free Machine Learning-based models to predict and avoid collision in mobile networks and resolve it before it occurs. Our models can be implemented in almost every mobile device, since they use only neighboring information to predict collision on top of that a GPS module is not required. The results demonstrate a promising potential of using Artificial Intelligence via Machine Learning modeling as a novel approach to avoid communication collision by using only neighboring information. Our models can achieve at least 70% accuracy for traffic in vehicular network in two-way highway, four-way intersection and four-way roundabout. Furthermore, they can avoid 80% of collisions in two-way highway and four-way intersection, as well as 65% in four-way roundabout. Which means we can decrease packet loss caused by collision by at least 65%, hence improve energy consumption.