Abstract

Constructing diagnostic Bayesian network models is a complex and time consuming task. In this thesis, we propose a methodology to simplify and speed up the design of very large Bayesian network models. The models produced using our methodology are based on two simplifying assumptions: (1) the structure of the model has three layers of variables and (2) the interaction among the variables can be modeled by canonical models such as the Noisy-MAX gate. The methodology is implemented in an application named GeNIeRate, which aims at supporting construction of diagnostic Bayesian network models consisting of hundreds or even thousands of variables. Preliminary qualitative evaluation of GeNIeRate shows great promise. The prediction is that GeNIeRate can reduce the model building time for an inexperienced Bayesian network model builder by 20-30%. We conducted an experiment comparing our approach to traditional techniques for building Bayesian network models by rebuilding a diagnostic Bayesian network model for liver disorders, HEPAR-II. We found that the performance of the model created with GeNIeRate is better than the performance of the original HEPAR-II.