Database mining is the nontrivial extraction of implicit, previously unknown, and potentially useful knowledge from a given database. Many researches have been conducted to find characteristic rules and discrimination rules. However there are still knowledge which are not included in proposed rules but might be useful.
In this thesis, we propose the Attribute-Value Dependency that searches discriminating attributes in databases to extend mining target. And we propose two measures called Attribute-Value Dependency Strength and Significance Factor to identify valid attribute-value dependencies. A canonical procedure for finding out attribute-value dependency is also presented with an illustrative example. Moreover, we adopt Attribute Concept Hierarchy as the domain knowledge in order to search more general rules.
Discovered attribute-value dependencies will be useful in understanding the characteristics of real-world application domain, along with other kinds of knowledge.
