In this paper, we describe virtual sensing framework (VSF), which reduces sensing and data transmission activities of nodes in a sensor network without compromising on either the sensing interval or data quality. VSF creates virtual
sensors (VSs) at the sink to exploit the temporal and spatial correlations amongst sensed data. Using an adaptive model at every sensing iteration, the VSs can predict multiple consecutive sensed data for all the nodes with the help of sensed data from a few active nodes. We show that even when the sensed data
represent different physical parameters (e.g., temperature and humidity), our proposed technique still works making it independent of physical parameter sensed. Applying our technique can substantially reduce data communication among the nodes leading to reduced energy consumption per node yet maintaining high accuracy of the sensed data. In particular, using VSF on the temperature data from IntelLab and GreenOrb data set, we have reduced the total data traffic within the network up to 98% and 79%, respectively. Corresponding average root mean squared error of the predicted data per node is as low as 0.36 °C and 0.71 °C, respectively. This paper is expected to
support deployment of many sensors as part of Internet of Things in large scales.