Applying Bayesian Networks to Sensor-Driven Systems

This paper discusses a middleware component, the likelihood estimation service (LES), that allows the application of Bayesian reasoning to a real sensor-driven environment. Using LES, first, a Bayesian network is learned from location data. Once trained, the network is used in order to estimate the likelihood of users´ spatio-temporal properties, such as the likelihood of their sighting in specific rooms. The learning algorithm is evaluated by calculating a confidence level. The output of the system is a first-order-logic predicate that is maintained in the SCAFOS middleware as approximate knowledge, even when the sensors fail.