Sidec: an Attribute{based Discovery System with Background Knowledge

The discovery of knowledge in databases is currently a very active research area. Many discovery systems adapt traditional attribute{based learners for the extraction of patterns. They have been, however, restricted by their inability to incorporate background knowledge into the learning process. In this paper an attribute{based learning algorithm, called SIDEC, which can incorporate a restricted form of Horn clauses as background knowledge is presented. SIDEC uses its background knowledge with the description of its training examples to augment the list of attributes used to describe the examples. It then uses CN2, an attribute{based learning algorithm, to produce a set of rules. This paper analyses the innuence of background knowledge into the learning process and shows that additional background knowledge can sometimes produced negative results. It is shown how SIDEC discovered useful patterns on a real database used by the Mexican electrical utility. The results varied considerably when diierent background knowledge was used.