Process-Based Database Support for the Early Indicator Method

In Wirth t Reinartz (1996), we introduced the early indicator method, a multi-strategy approach for the efficient prediction of various aspects of the fault profile of a set of cars in a large automotive database. While successful, the initial implementation was limited in various ways. In this paper, we report recent progress focusing on performance gains we achieved through proper process-based database support. We show how intelligent management of the information collected during a KDD process can both make the task of the user easier and speed up the execution. The central idea is to use an object oriented schema as the central information directory to which data, knowledge, and processes can be attached. Furthermore, it enables the automatic exploitation of previously stored results. Together with the query-shipping strategy, this achieves efficiency and scalability in order to analyze huge databases. While we demonstrate the usefulness of our solution in the context of the early indicator method, the approach is generally applicable and useful for any integrated, comprehensive KDD system.