의사결정나무의 광역 최적화를 이용한 분류 방법

Abstract

Classification is to generate a set of rule of classifying objects into severalcategories based on the training sample. Decision tree as a classification tool is being usedsuccessfully, because it has the advantage of being a knowledge representation intuitivelycomprehensible to the user. In this paper, we propose a new classification usingoptimization of decision tree. The proposed method consists of three phases. First, wechoose the relevant variables using a well-known decision tree algorithm, classificationand regression tree(CART). Second, we find the optimum thresholds simultaneouslyusing adaptive particle swarm optimization(APSO) for those selected variables. Third, wesimplify the set of IF-THEN rules. Additionally, we repeat the procedure to find moreimproved rules by changing the splitting variables. To validate the proposed method,several artificial and real datasets are used. We compare our results with the originalCART results and show that the proposed algorithm is promising for improvingprediction accuracy.