CAJ | 학술논문

动态分类器集成选择(DCES)是当前集成学习领域中一个非常重要的研究方向。然而，当前大部分 DCES算法的计算复杂度较高。为了解决该问题和进一步提高算法的性能，本文提出了基于聚类的动态分类器集成选择(CDCES)，该方法通过对测试样本聚类，极大地减少了动态选择分类器的次数，因而降低了算法的计算复杂度。同时， CDCES是一种更加通用的算法，传统的静态选择性集成和动态分类器集成为本算法的特殊情况，因而本算法是一种鲁棒性更强的算法。通过对UCI数据集进行测试，以及与其他算法作比较，说明本算法是一种有效的、计算复杂度较低的方法。
동태분류기집성선택(DCES)시당전집성학습영역중일개비상중요적연구방향。연이，당전대부분 DCES산법적계산복잡도교고。위료해결해문제화진일보제고산법적성능，본문제출료기우취류적동태분류기집성선택(CDCES)，해방법통과대측시양본취류，겁대지감소료동태선택분류기적차수，인이강저료산법적계산복잡도。동시， CDCES시일충경가통용적산법，전통적정태선택성집성화동태분류기집성위본산법적특수정황，인이본산법시일충로봉성경강적산법。통과대UCI수거집진행측시，이급여기타산법작비교，설명본산법시일충유효적、계산복잡도교저적방법。
Dynamic classifier ensemble selection (DCES) is an important field in the machine learning. However, computational complexity of the current methods is very high. In order to solve the problem and improve the performance further, cluster based dynamic classifier ensemble selection (CDCES) is proposed in this paper. Using the proposed method to cluster the testing sample, the degree of DCES is reduced enormously and the computation complexity is decreased. At the same time, CDCES is a more general method and the traditional static ensemble selection and dynamic classifier is a specific case of the proposed method, so CDCES is more robust. Compared with the other algorithms on UCI data set, it is demonstrated that the proposed method is a more effective and lower computational complexity method.