ACO-based BN Structure Learning And Its Application

In AI, uncertainty reasoning has been a focus of research. Bayesian Network is natural compact graphical representation of joint probability distribution, which can express a potential dependent relationship among uncertain variables and can exploit the structure of the domain. Because of its merits, Bayesian Network has been a powerful tool to solve many uncertainty problems, and become a mainstream within the AI probabilistic and uncertainty community.The first task of applying BN is construction. Nonetheless, it is often difficult, subjective and time-consuming to construct BN from expert knowledge alone, particularly because of the need to provide numerical parameters. Methods for capturing available data to construct BN or refine an expert-provides network promise to greatly improve both the efficiency of knowledge engineering and the accuracy of the methods.In this paper, we learn BN structure from complete data. On the basis of the research and analysis of the current BN structure learning algorithms, we introduce ACO algorithm and I<2 metric into score-based BN structure learning algorithm. The details are given as follows.(1) ACO-based BN structure learning algorithm: We use ACO algorithm as search procedure and K2 metric as score metric. The search space in which the ACO algorithms operate can be defined in two different ways: ordering and dag. We propose Ant ordering algorithm in ordering and ACO B algorithm in dag. We discuss the different variants of ACO B algorithm in this paper.(2) BN application in CRM: BN is applied to the part of customer analyzing (Data mining) in CRM. We use BN into eliciting useful information from customer database in order to satisfying the needs of the decision-maker.