摘要
研究了基于偶对约束的半监督模糊聚类,将马氏距离引入到半监督模糊聚类SCAPC(semi-supervised fuzzy clustering algorithm with pairwise constraints)中,获得了一种新的半监督模糊聚类目标函数,通过求解优化问题,提出了一种基于偶对约束和马氏距离的半监督模糊聚类算法M-SCAPC(ModifiedSCAPC).针对选择的标准数据集和人工数据集,对提出的算法M-SCAPC进行了实验研究,并与FCM(fuzzy C-means)、AFCC(active fuzzy constrained clustering)和SCAPC算法的聚类性能进行了比较,表明了提出的算法M-SCAPC在收敛速度和正确率方面的有效性.
The semi-supervised fuzzy clustering based on pair-wise constraints which introduces Mahalanobis distance into SCAPC(Semi-supervised fuzzy Clustering Algorithm with Pairwise Constraints)algorithm is mainly studied.And a new semi-supervised fuzzy clustering objective function is obtained.By solving the optimization problem,a semi-supervised fuzzy clustering algorithm M-SCAPC (Modified SCAPC) based on pairwise constraints and Mahalanobis distance is proposed.And some experimental researches are conducted for M-SCAPC algorithm using the selected standard data set and the artificial data set.Besides,clustering performance on M-SCAPC algorithm are compared with that of FCM(Fuzzy C-Means),AFCC (Active Fuzzy Constrained Clustering) and SCAPC algorithms.From the results,M-SCAPC is effective in the convergence speed and the accuracy.
出处
《河北大学学报(自然科学版)》
CAS
北大核心
2014年第5期535-540,共6页
Journal of Hebei University(Natural Science Edition)
基金
国家自然科学基金资助项目(61375075)
河北省自然科学基金资助项目(F2012201014)
关键词
半监督聚类
偶对约束
度量学习
马氏距离
semi-supervised clustering
pairwise constraints
metric learning
mahalanobis distance