Using Knowledge Discovery Techniques for Database Schema Refinement

This paper proposes the automated translation of rules extracted from data mining or knowledge discovery tools into active database rules. We term this process of translating a knowledge discovery rule and incorporating it into a database schema in the form of an ECA (eventcondition-action) rule as database schema refinement. We introduce a new rule identification measure for categorising knowledge discovery rules into semantic integrity constraints and probabilistic rules, which forms the basis for determining the mode of constraint enforcement in the enriched schema. This measure is based on database statistics and history. Using this measure we estimate and model both the static and dynamic characteristics of a given rule. The rule classification process is followed by the generation of the ECA equivalents. We present a generic technique and show how this is implemented in an Oracle database. Finally, the monitoring of the database scheme refinement process and the transition of a rule from one state to another are discussed.