摘要
学习器间的差异性是影响集成学习效果的一个关键因素。目前针对分类集成的研究较多,针对聚类集成的研究则相对较少。基于聚类问题的本质特点,提出一种新的聚类集成学习方法,利用聚类有效性指标度量不同聚类结果性能上的差异,根据有效性指标的评价值为聚类结果分配权值,通过加权投票的决策方法进行聚类集成并确定最佳聚类数。理论研究和实验结果证明了新的聚类集成学习方法的可行性和高效性。
The diversity of learners is an important factor to the performance of ensemble learning.Until now,a large amount of researches in classified ensembles has been done,while limit work focuses on the clustering ensemble.Based on the essential characteristic of clustering,this paper proposes a new method of clustering ensemble learning,which measures the diversity in performance of different clustering results by using cluster validity index.The weights for each cluster result are constituted according to the validity index evaluation.And then the clustering ensemble is conducted and the optimal clustering number is determined both by the decision-making approach of weighted-voting.Theoretical research and experimental results demonstrate the feasibility and high efficiency of the new clustering ensemble leaning method.
出处
《计算机应用与软件》
CSCD
北大核心
2012年第9期45-49,70,共6页
Computer Applications and Software
基金
国家自然科学基金项目(61139002)
中国民用航空局科技项目(MHRD201006
MHRD201101)
关键词
聚类
集成学习
有效性指标
差异性
Clustering ,Ensemble learning ,Validity index ,Diversity
