Rule induction and instance-based learning a unified approach

摘要: This paper presents a new approach to inductive learning that combines aspects of instancebased and rule induction in single simple algorithm. The RISE system searches for rules specific-to-general fashion, starting with one per training example, avoids some the difficulties separate-and-eonquer approaches by evaluating each proposed step globally, i e, through an efficient procedure is equivalent checking accuracy set as whole on every example. Classification performed using best-match strategy, reduces nearest-neighbor if all generalizations instances were rejected. An extensive empirical study shows consistently achieves higher accuracies than state-of-the-art representatives its "parent" paradigms (PEBLS CN2), also outperforms decision-tree learner (C4 5) 13 out 15 test domains (in 10 95% confidence).