The use of Bayesian networks to support context aware mobile devices.

Mobile communications technology has increasingly become integrated into modern day life and the rate of this integration is growing rapidly. However, many mobile devices including personal digital assistants (PDA) and mobile phones are predominantly passive, lacking a facility for automated information handling. For example, a mobile phone cannot automatically detect a user's context and react to support that situation. In this thesis, we present a context aware system that automatically detects mobile users' activity context through information gathered from a mobile phone. The thesis contains two main contributions: the development of the context prediction model and the implementation of a context aware system. The development of the context prediction model involves the use of Bayesian networks to solve the uncertainty regarding a user current activity, along with the prediction of possible future activities. A Bayesian probabilistic network is a comprehensive probabilistic computational framework for reasoning under conditions of uncertainty. In this research, we utilise Bayesian networks as a modelling tool to support context reasoning and activity prediction within the context aware domain. A further contribution of this work is the development of a real world context aware system based upon our above theoretical contribution. A Sony Ericsson P910i mobile phone alongside a context aware engine server has been implemented to demonstrate the theoretical principles. The mobile phone is used to capture contextual information from a user and interact with the context engine server. Using a probabilistic approach and uncertainty metric, the Bayesian context prediction network (which is located on the context engine server) predicts a user's likely activity, and contextual information is provided to the user via their mobile phone. A dataset of real-life user activities has been collected to evaluate the accuracy of the predicted activities. The results of the evaluation demonstrated encouraging levels of accuracy for this approach