Multiagent Bayesian forecasting of time series
Tune series are used in a wide variety of fields including engineering, science, sociology, and economics. Forecasting of time series allows for useful inferences where agents are able to estimate the future state of a problem domain. In this thesis, we study multiagent forecasting of time series. We describe a dynamic multiply sectioned Bayesian network (DMSBN) framework, that is used by our agents to cooperatively perform forecasting. An algorithm is presented which allows our agents to perform the forecast with exact probabilistic inference. We show superior, performance of our agents over agents using dynamic Bayesian networks (DBNs) through experiment.