Predicting link quality using supervised learning in wireless sensor networks

Abstract

Routing protocols in sensor networks maintain information on neighbor states and potentially many other factors in order to make informed decisions. Challenges arise both in (a) performing accurate and adaptive information discovery and (b) processing/analyzing the gathered data to extract useful features and correlations. To address such challenges, this paper explores using supervised learning techniques to make informed decisions in the context of wireless sensor networks. We investigate the design space of both offline learning and online learning and use link quality estimation as a case study to evaluate their effectiveness. For this purpose, we present MetricMap, a metric-based collection routing protocol atop MintRoute that derives link quality using classifiers learned in the training phase, when the traditional ETX approach fails. The offline learning approach is evaluated on a 30-node sensor network testbed, and our results show that MetricMap can achieve up to 300% improvement over MintRoute in data delivery rate for high data rate situations, with no negative impact on other performance metrics. We also explore the possibility of using online learning in this paper.