Flexibly Exploiting Prior Knowledge in Empirical Learning

Abstract

This paper presents a method to incorporate knowledge from possibly imperfect models and domain theories into inductive learning of decision trees for classification. The approach assumes that a mode l or domain theory reflects useful prior knowledge of the task. Thus , the default bias should accept the model's predictions as accurate, even in the face of somewhat cont radictory data, which may be unrepresentative or noisy. However , our approach allows the system to a bandon the model or domain theory, or portions thereof, in the face of sufficiently contradictory data . In particular, we use C4 .5 to induce decision t rees from data that hav e been augmented by model or domaint heory-derived 'features ' . We weakly bias the system to select model-derived features during decision tree induction, but this preference is not dogmatically applied. Our experiments vary imperfection in a model, the representativeness of data , and th e veracity with which model-derived f eat ures are preferred .