Toward reliable data analysis for Internet of Things by Bayesian dynamic modeling and computation

In this paper, a Bayesian dynamic model is proposed to evaluate the sensor nodes´ credibilities online, in a paradigm of agricultural Internet of things (IoT). The purpose is to discriminate reliable and unreliable data items before further data analysis, and thus to implement reliable data analysis. The credibility of the sensor node of interest is treated as the state variable of the model. The proposed model is composed of a state transition function, which characterizes the time-varying property of trustworthiness, and a likelihood function, which connects the state variable with the sensor measurements. A voting mechanism employing measurements of neighbor nodes is used to construct the likelihood function. Based on the model, the Bayesian rule is performed for statistical inference on the sensor´s credibility, the whole information of which is encoded in the posterior density function. Due to a nonlinear form of the model, there is no closed form solutions to calculate the posterior. So a particle filtering method is chosen to approximate the posterior online. The efficiency of the proposed model is verified by numerical simulations.