Learning in an effective way makes enterprises more competitive in the rapidly changing world. For the purpose of improving learning effectiveness, the combination of user knowledge and statistical data has become an important issue in Bayesian network (BN)-related research. This research is focused on the relevance between random variables, and it describes an approach for learning BNs that combines the user's knowledge and statistical data. A BN is a graphical model that encodes the probabilistic relationships among a set of variables. Such a network is a knowledge representation tool, and it is based on the new information that is obtained, which easily can be extended or used to reduce the network to fit the changing knowledge. However, in order to construct a BN, heavy reliance on domain knowledge is required. To address this challenge, in this research, we propose iterative techniques for effective and efficient Bayesian structure learning. We used an entropy-based, link-strength technique that represents what the relationships between random variables are believed to be. Also, we provide a homogeneity test to ensure the quality of the learned structure. The results of this study are useful for the learning required for decision support systems. ?2013 Chinese Institute of Industrial Engineers.
Relation:
Journal of Industrial and Production Engineering,Vol.30,Issue3,P.144-159
