Context classifier for position-based user association control in vehicular hotspots

Unintentional associations of mobile devices to on-board WiFi access points (APs) can affect the outdoor Internet experience of mobile device users, as the on-going cellular connection is broken and a short-lived WiFi connection is initiated. This disruption of the user experience can be avoided if the on-board AP learns whether the user device is inside or outside the bus and decides to accept its connection request or not. In this article, we present a classifier-based mechanism for on-board APs that accepts or denies user device associations based on a classification of the relative position of the device. An analysis of the problem in terms of connection duration and RSSI is presented to motivate the selected approach. We then describe a classifier to identify the user relative position trained on features extracted from contextual information. The classifier was trained with a large dataset of real-world WiFi-usage and mobility patterns of a public bus fleet from Porto, Portugal. The training procedure indicated bus speed as the most relevant feature, and that the RSSI measured at the on-board AP does not contribute. Finally, we propose a mechanism that grants or denies connection access to users based on the classifier output. We discuss how to integrate this mechanism in the AP network stack and evaluate its performance in real-world tests. Our solution can avoid 40% of the associations from users outside of the bus.