|
Knowledge Engineering Bayesian Networks (KEBN) |
||
|
Navigation:
Background |
Knowledge engineering is the process of building expert systems, such as Bayesian Networks, for application domains. It involves investigating an application domain, identifying important concepts and expressing these concepts as objects and relationships in the formal representation of the expert system. The knowledge engineer is, usually, either trained in the formal representation of the expert system or is an expert in the application domain, but often not in both. For this reason the knowledge engineering task can be seen to address two issues. The first, how can someone, especially someone who isn't an expert in the application domain, properly identify the important concepts? And second, how can someone, especially someone who isn't trained in the formal representation of the expert system, express the concepts that they have identified as important? To address these issues the field of knowledge engineering emerged, its objective, to formalize the process of building expert systems, ensuring they are created correctly and used to their maximum potential. Bayesian Networks (BNs) are graphical expert systems for reasoning with probabilities. BNs are used to identify the posterior probability of an event given observations of current system state. BN's are composed of ontological, qualitative and quantitative components [1, Chapter 3]. The ontological component is represented by a set of variables, also referred to as nodes, which can take on various values, also referred to as states. The qualitative component is represented by a graphical structure composed of nodes with directed links representing causal influences between parent and child variables. The quantitative component is represented by conditional probability tables (CPTs) which quantify the effects of causal variables. Bayes' theorem is central to the inference mechanisms used to update the posterior probabilities of the variables using the probabilistic information of the CPTs and dependency information of the casual structure. When developing BNs, knowledge engineering techniques need to be applied to the development of each of these components, separately and collectively. Although there is no strict necessity to use formal knowledge engineering techniques, experience has shown that these tasks can be extremely difficult, especially when modeling complex systems. As the theory of BNs is reasonably well understood the emphasis of current research involving BNs is on overcoming their development difficulties, via the development of better support tools, and investigating their applicability to certain domain areas. The modeling shell Netica (Norsys, Inc.) was used to construct the BNs identified in this thesis. |