Functional Dependency Discovery via Bayes Net Analysis

Functional dependency plays a key role in database normalization, which is a systematic process of verifying database design to ensure the nonexistence of undesirable characteristics. Bad design could incur insertion, update, and deletion anomalies that are the major cause of database inconsistency. In this paper, we propose a novel technique to discover functional dependencies from the database table. The discovered dependencies help the database designers covering up inefficiencies inherent in their design. Our discovery technique is based on the structure analysis of Bayesian network or Bayes net. Most data mining techniques applied to the problem of functional dependency discovery are rule learning and association mining. Our work is a novel attempt of applying the Bayes net to this area of application. The proposed technique can reveal functional dependencies within a reduced search space. Therefore, computational complexity is acceptable. Key-Words: Functional dependency discovery, Bayesian network, Data mining, Database normalization.