iOOBN: An Object-Oriented Bayesian Network Modelling Framework with Inheritance

This thesis presents an OOBN framework, iOOBN, that extends existing frameworks of OOBNs to include inheritance, abstract classes and interfaces. We demonstrate how iOOBN can be used to build decision-making applications for real-life problems with uncertainty and present a prototype implementation of the framework. Then we evaluate both the framework and implementation by re-engineering a real-life project and several existing OOBNs. In the thesis, we also present a new incremental compilation algorithm for OOBNs. This algorithm allows the efficient compilation of OOBNs without them being flattened. Finally, we describe an algorithm that automatically learns the class hierarchies of OOBNs. This algorithm can be used to maximise re-use and produce better constructions of class inheritance hierarchies.