摘要
传统谱聚类对初值选取十分敏感,严重影响了聚类效果。为了解决初值敏感问题,提出了基于CMTFCM(借鉴历史知识的类中心距离极大化聚类算法)的自适应谱聚类算法。该算法以样本空间的标准差作为尺度参数,实现了尺度参数的自适应选取,提高了算法效率;而通过借鉴历史知识,引入类中心距离极大化项,避免了干扰点对类中心的干扰,提高了算法鲁棒性。通过在模拟数据集以及真实数据集上的测验,取得了比传统谱聚类更稳定的聚类效果,验证了算法的有效性。
The traditional spectral clustering is sensitive to the selection of initial value,seriously affecting the clustering effect. In order to solve the problem,this paper proposed the adaptive spectral clustering algorithm based on CMT-FCM( transfer fuzzy C-means clustering algorithm with center distance maximization). This algorithm used the standard deviation of sample space as the scale parameter,which realized the adaptive selection of the scale parameter,improved the efficiency of the algorithm. By the guidance of historical knowledge and introducing center distance maximization,the algorithm avoided the problem that interference information had a certain appeal to every class center,improving the robustness of the algorithm. The experimental studies on both artificial and real-world datasets achieve more stable clustering results than the traditional spectral clustering. It demonstrates the effectiveness of the algorithm.
出处
《计算机应用研究》
CSCD
北大核心
2017年第3期739-742,共4页
Application Research of Computers
基金
江苏省自然科学基金重点研究专项资助项目(BK2011003)
关键词
谱聚类
初值敏感
标准差
历史知识
中心距离极大化
spectral clustering
sensitivity of initial value
standard deviation
historical knowledge
center distance maximization