E-viri
-
Wang, LiMin; Yao, GuoFeng; Li, XiongFei
International Information Institute (Tokyo). Information, 01/2012, Letnik: 15, Številka: 1Journal Article
How to combine functional dependencies of relational database into probabilistic model learning procedure is a tough problem for knowledge discovery. This paper presents a novel Bayesian model named NB-TAN (NT), which utilizes the conditional independence assumption of Naive Bayes while trying to maintain the inter-dependencies between attributes. Thus a good tradeoff between model complexity and learnability can be realized in practice. Canonical cover is utilized to divide attributes into several related groups and extraneous attributes can be found. This helps to reduce the computational complexity and improve the robustness of the classifier. PUBLICATION ABSTRACT
