Utilising domain knowledge in inductive knowledge discovery

Describes a novel inductive knowledge discovery and learning algorithm, CUPID, and how it is able to utilise basic forms of domain knowledge to guide and improve its efficiency and accuracy. The system is capable of using two forms of domain knowledge-generalisation hierarchies on attribute values, and constructed intensional attribute definitions. Experimental results have been gathered suggesting that the use of these types of domain knowledge is beneficial in induction-by allowing more efficient computation and more accurate and predictive resulting knowledge structures. A well-founded information theoretic measure of rule utility (the J measure) allows effective pruning ofthe hypothesis search space. CUPID´s inductive strategy and information-theoretic guided search of the hypothesis space allows available domain knowledge that may be incorrect and incomplete to be utilised