Lightweight Sensing Uncertainty Metric—Incorporating Accuracy and Trust

The future will involve millions of networked sensors whose sole purpose is to gather data about various phenomena so that it can be used in making informed decisions. However, each measurement performed by a sensor has an associated uncertainty in its value, which if not accounted for properly, could potentially derail the decision process. Computing and embedding the associated uncertainties with data are, therefore, crucial to providing reliable information for sensor-based applications. In this paper, we present a novel unified framework for computing uncertainty based on accuracy and trust. We present algorithms for computing accuracy and trustworthiness and also propose an approach for propagating uncertainties. We evaluate our approach functionally by applying it to data sets collected from past deployments and demonstrate its benefits for in-network processing as well as fault detection.