基于灰关联分析的加权自适应谱聚类算法

摘要

为了降低谱聚类采用高斯函数作为相似性度量方式对参数的敏感性，以及能够发现多密度簇的同时降低噪声点的干扰，提出了一种将基于均衡接近度的灰关联分析结合到谱聚类中的新方法，采用加权的自适应相似性度量方式。最后用改进的FCM算法对其进行聚类。在真实数据集和人工数据集上分别对提出的算法和现有算法进行了比较分析。研究结果表明，提出的新算法能够消除参数的影响，具有更高的聚类精度。聚类精度采用F测度指标。%Aiming to reduce sensitivity to parameter by using gaussian function as the similarity measure in spectral clustering and to identify multidimensional clusters with reducing the outliers in-terference, a new approach is presented, in which grey relational analysis based on the balanced closeness degree is integrated with spectral clustering and an adaptive weighted similarity measure is used, and finally the improved FCM algorithm is used for clustering.The proposed algorithm is com-pared with the existing algorithms on the artificial and real-world datasets.The experimental results demonstrate that the proposed algorithm can avoid the effect of parameter and achieve higher cluste-ring accuracy by using F measurement index.