A Communication-efficient Framework For Outlier-free Data Reporting In Data-gathering Sensor Networks

摘要

In this paper, we address the problem of reducing the communication cost and hence the energy costs incurred in data-gathering applications of a sensor network. Environmental data depicts a huge amount of correlation in both the spatial and temporal domains. We exploit these temporal-spatial correlations to address the aforementioned problem. More specifically, we propose a framework that partitions the physical sensor network topology into a number of feature regions. Each sensor node builds a data model that represents the underlying structure of the data. A representative node in each feature region communicates only the model coefficients to the sink, which then uses them to answer queries. The temporal and spatial similarity has special meaning in outlier cleaning too. We use a modified 2-score technique to precisely label the outliers and use the spatial similarity to confirm whether the outliers are due to a true change in the phenomenon under study or due to faulty sensor nodes.