Multi-relational Structural Bayesian Classifier

In the traditional naive Bayes classification method, training data are represented as a single table (or database relation), where each row corresponds to an example and each column to a predictor variable or a target variable. In this paper we propose a multi-relational extension of the naive Bayes classification method that is characterized by three aspects: first, an integrated approach in the computation of the posterior probabilities for each class; second, the applicability to both discrete and continuous attributes, third, the consideration of knowledge on the data model embedded in the database schema during the generation of classification rules. The proposed method has been implemented in the new system Mr-SBC and tested on three benchmark tasks. Results on predictive accuracy favour our system for the most complex task. Mr-SBC also proved to be an efficient multi-relational data mining system with a tight dose integration to a relational DBMS.