A Bayesian-networks-based approach for managing uncertainty in location-tracking applications

In this paper we present a probabilistic model for managing uncertainty in location-tracking applications. Our technique tries to capitalize on the fact that humans are creatures of habit, which makes it possible to predict how a person would react in a certain situation, based on his/her history. Since the moving objects (MO) we are interested in here are humans, we refer to this subset of MO as roving users or RU. Our approach is based on using Bayesian networks to model the RU behavior and preferences. This user model is then used to predict his/her actions allowing the system to more accurately determine his/her location between location reports or in the near future. Our technique estimates the location of a RU based on the last location reported, the user model, and knowledge about the current state of affairs. We present the results of a simulation we performed comparing the accuracy of location estimation of a popular technique to that of our proposed probabilistic technique.