PRM-based multi-relational association rule mining

One of the primary goals of data mining is to extract patterns from a large volume of data. Rules characterize patterns in a humanly comprehensible manner. In particular, association rules have become one of the central research topics in data mining. Association rule mining has previously been restricted to data in a single table. As most data-intensive applications employ a relational database for storage and retrieval, this thesis aims at mining association rules from a standard relational database. Fundamentally different from previous works, the proposed algorithm is driven by the Probabilistic Relational Model (PRM) of a relational database rather than the minimum support restriction. Based on the conditional independence relationships inferred from the PRM structure, our algorithm removes a set of antecedent attributes that lead to the generation of redundant rules to improve the learning efficiency and effectiveness.