Bayesian Belief Networks for Predictive Modeling and Decision Support

Bayesian belief networks (BNs) are models that graphically and probabilistically represent associations among variables. BNs have been shown to be efficient representation tools for modeling and describing domains containing some degree of uncertainty due to an imperfect understanding of the underlying states. The resulting knowledge representations enable visual depiction of complex stochastic systems and serve as the basis for the development of efficient inference algorithms. BNs are increasingly being used as the model base in knowledge based systems. Some examples of applications include predictive modeling, decision support, risk analysis, diagnostics, and troubleshooting. Enhancements in computing capabilities as well as algorithms have made possible the development and evaluation of BNs using personal computing platforms. An overview will be presented for some open source and proprietary software applications used for BN modeling and inference. Examples will be presented for the application of BNs in agricultural research. Advantages and disadvantages will be outlined for using BNs in agricultural research. A worked example will be given for the development of a Bayesian belief network from empirically derived data.