One-Pass Learning Algorithm for Fast Recovery of Bayesian Network

All Papers

An efficient framework is proposed for the fast recovery of Bayesian network classifier. A novel algorithm, called Iterative Parent-Child learning-Bayesian Network Classifier (IPC-BNC), is proposed to learn a BNC without having to learn the complete Bayesian network first. IPC-BNC was proved correct and more efficient compared with a traditional global learning algorithm, called PC, by requiring much fewer conditional independence (CI) tests. Besides, we recognize and introduce AD-tree into the implementation so that computational efficiency is further increased through collecting full statistics within a single data pass. The IPC-BNC and AD-tree combination is demonstrated very efficient in time by our empirical study, making itself an attractive solution in very large applications. Keywords: Bayesian Network classifier, IPC-BNC, AD-tree