A Technique for Advanced Dynamic Integration of Multiple Classifiers

Currently electronic data repositories are growing quickly and contain huge amount of data from commercial, scientific, and other domain areas. Knowledge discovery in databases (KDD) is an emerging area that considers the process of finding previously unknown and potentially interesting patterns and relations in large databases. Most current KDD systems offer only isolated discovery techniques, and very few systems use a combination of the available discovery techniques. Our goal is to design an architecture of an integrated knowledge discovery management system (IKDMS), which enables integration of multiple discovery techniques forming a platform upon which different KDD applications can be build. In this paper our focus is on the method evaluation/selection subsystem of an IKDMS. This subsystem is very important in any IKDMS because it helps a user to select an appropriate data mining method among the supported ones. We present and evaluate a technique for advanced dynamic integration of multiple classifiers that is based on the assumption that each classifier is the best only inside certain sub domains of the whole application domain. We have made experiments using three databases included in the University of California Machine Learning Repository achieving promising results either in diagnosis accuracy or in the time requirements of diagnostics or both.