Feature transformation through rule induction : A case study with the k-NN classifier
Antal van den Bosch · Research portal (Tilburg University) · 2004
Abstract. A extension to the k-nearest neighbor classifier is described in which automatically induced rules are used as binary features, which are active in an instance when the left-hand side of the corresponding rule matches with the instance. The ripper rule induction algorithm is employed to produce the rules. The similarity between a memory instance and a new instance is based on the rules the two instances share. We report on experiments that indicate that (i) the method equals the generalization performances of ripper and k-NN classification on average, and (ii) when the original multi-valued features are combined with the transformed rule-based features, some significant improvements in k-NN classification are observed, particularly with artificial benchmark tasks. 1 Rules as features A well-established machine-learning solution to classification problems is rule induction [1–3]. The goal of rule induction is generally to induce a set of rules