摘要
Leader算法是一种基于粗糙集的层次聚类算法,其聚类过程中各维特征贡献作用均等,这样的处理方法降低了主要特征在聚类中的贡献作用,从而影响聚类的效果.采用负投影梯度法对各维特征的权重进行自适应学习,从而优化地获得各维特征的权值,进而建立了基于负投影梯度法的特征权重Leader聚类算法.该算法强化了重要特征在聚类过程中的作用.聚类结果用"熵"和"精度"来评价,实验结果证明,改进后的聚类算法能够改善聚类的效果,验证了本文方法的可行性与有效性.
Leader algorithm is a kind of hierarchical clustering algorithm for large data sets based on rough set. The weakness of Leader is that each feature has the same contribution for the clustering process. If we can enlarge the contribution of important features, we can improve the effect of the cluster. In this paper, the negative projection gradient method is adopted to calculate the each featureweight adaptively. Therefore we propose an algorithm named Feature-Weighting Leader based on negative gradient projection (FWLeader) which can project the original feature space into new space according to the weights. FWLeader can enhance the contribution of important attributes in the process of clustering. Entropy and precision are used to evaluate the performance of clustering algorithm. Experimental results show that the improved clustering algorithm can improve the performance of Leader.
出处
《小型微型计算机系统》
CSCD
北大核心
2014年第9期2147-2150,共4页
Journal of Chinese Computer Systems
基金
河北省自然科学基金项目(F2013202138
H2012202035)资助
河北省教育厅重点项目(ZH2012038)资助
河北省高等学校青年基金项目(SQ121006)资助