摘要：There are two categories of well-known approach (as basic principle of classification process) for learning structure of Bayesian Network (BN) in data mining (DM): scoring-based and constraint-based algorithms. Inspired by those approaches, we present a new CB* algorithm that is developed by considering four related algorithms: K2, PC, CB, and BC. The improvement obtained by our algorithm is derived from the strength of its primitives in the process of learning structure of BN. Specifically, CB* algorithm is appropriate for incomplete databases (having missing value), and without any prior information about node ordering.
