Toward Effective Knowledge Acquisition with First-Order Logic Induction

Knowledge acquisition with machine learning techniques is a fundamentalrequirement for knowledge discovery from databases and data miningsystems. Two techniques in particular --- inductive learning andtheory revision --- have been used toward this end. A method thatcombines both approaches to effectively acquire theories (regularity)from a set of training examples is presented. Inductive learning is usedto acquire new regularity from the training examples; and theoryrevision is used to improve an initial theory. In addition, a theorypreference criterion that is a combination of the MDL-based heuristicand the Laplace estimate has been successfully employed in the selectionof the promising theory. The resulting algorithm developed byintegrating inductive learning and theory revision and using thecriterion has the ability to deal with complex problems, obtaininguseful theories in terms of its predictive accuracy.