Seminar on Computational Learning and Adaptation

K-Optimal Rule Discovery: An Efficient and Effective Approach to Exploratory Data Mining

Geoff Webb
Faculty of Information Technology
Monash University

Most data-mining techniques seek a single model that optimizes an objective function with respect to the data. In most real-world applications several models will equally optimize this function, although they may not all satisfy a user's preferences. Thus, the program will make an arbitrary decision, whereas users may base their selections on background knowledge and pragmatic considerations that are infeasible for the automated system to quantify. Methods for exploratory rule discovery, of which association rule discovery is the best known example, seek all models that satisfy user-defined criteria. This lets the user select between these models rather than relinquishing control to the program. K-Optimal rule discovery is an exploratory technique which finds the k rules that optimize a user-selected objective function while respecting other user-specified constraints. I argue that this approach has important advantages over the minimum-support technique that underlies association-rule discovery and I present efficient algorithms that support it.

Return to the seminar schedule