A MULTISTRATEGY APPROACH TO RELATIONAL KNOWLEDGE DISCOVERY IN DATABASES

When learning from very large databases, the reduction of complexity i s extremely important. Two extremes of making knowledge discovery in d atabases (KDD) feasible have been put forward. One extreme is to choos e a very simple hypothesis language, thereby being capable of very fas t learning on real-world databases. The opposite extreme is to select a small data set, thereby being able to learn very expressive (first-o rder logic) hypotheses. A multistrategy approach allows one to include most of these advantages and exclude most of the disadvantages. Simpl er learning algorithms detect hierarchies which are used to structure the hypothesis space for a more complex learning algorithm. The better structured the hypothesis space is, the better learning can prune awa y uninteresting or losing hypotheses and the faster it becomes. We hav e combined inductive logic programming (ILP) directly with a relationa l database management system. The ILP algorithm is controlled in a mod el-driven way by the user and in a data-driven way by structures that are induced by three simple learning algorithms.