CAJ | 학술논문

　　传统决策树算法有好的可理解性，而支持向量机集成方法有好的分类性能。本文基于二者的优点提出了一种基于支持向量机集成的决策树分类算法，首先在样本集合上训练支持向量机集成分类器。然后随机生成一定数目的特征向量并加以标注，构造辅助样本。最后在辅助样本的帮助下生成新的决策树。由于最终结果由决策树给出，所以本文算法具有很好的可理解性。此外在辅助样本的帮助下，算法的分类性能也得到提高。在UCI标准数据集与新闻文本数据集上的实验充分验证了本文算法的合理性。
　　전통결책수산법유호적가리해성，이지지향량궤집성방법유호적분류성능。본문기우이자적우점제출료일충기우지지향량궤집성적결책수분류산법，수선재양본집합상훈련지지향량궤집성분류기。연후수궤생성일정수목적특정향량병가이표주，구조보조양본。최후재보조양본적방조하생성신적결책수。유우최종결과유결책수급출，소이본문산법구유흔호적가리해성。차외재보조양본적방조하，산법적분류성능야득도제고。재UCI표준수거집여신문문본수거집상적실험충분험증료본문산법적합이성。
Traditional decision tree algorithm has good comprehensibility, and support vector machine integration ap-proach has good classification accuracy. Based on their advantages, we propose a support vector machine ensemble-based decision tree algorithm. Firstly, train a support vector machine ensemble classifier on training set. Then compute the labels of feature vectors to construct auxiliary samples. Finally, we train a decision tree under the help of auxiliary samples. Since the results are given by decision tree, the algorithm has good comprehensibility. Moreover, the perform is also improved with the help of the auxiliary samples. Experiments on UCI standard data sets and news text data set fully explain the reasonableness of the proposed algorithm.