Machine Learning for Resources Prediction in Multihoming Scenarios

Nowadays, mobile terminals have the ability to connect to several access networks at the same time. In an IoT scenario, it is important that all technologies be available simultaneously to transfer the information between all (users, sensors, vehicles) and the infrastructure. This paper proposes the use of machine learning to gather accurate network information in a real multihoming environment. It predicts the network information and, consequently, reduces the overhead of intrusive measurement processes. We propose a learning mechanism that extracts the required information, creates its own database in a dynamic way, and identifies when the existing information is enough to perform a "good" prediction. Due to its characteristics, this can be seamlessly adapted to different scenarios. The results obtained in a real scenario demonstrate that this approach significantly reduces the use of an intrusive measurement approach (by around 80%), while keeping the accuracy of the information and the performance of multihoming.