基于数据加权策略的模糊聚类改进算法

Improved Fuzzy Clustering Algorithm Based on Data Weighted Approach

该文提出了一种数据指数加权的模糊均值聚类策略,引入了指数权因子和影响指数,使得可以在聚类过程中差异化处理各个数据。新策略和现有的Gustafson-Kessel(G-K)算法相结合,提出了一种新的模糊聚类算法DWG-K用于提高聚类质量和挖掘离群点。数据试验表明DWG-K在提高聚类质量方面优于现有的G-K;在离群点挖掘方面,DWG-K对离群点的判定是全局的,离群点的物理意义清楚,且计算效率明显高于当前广泛采用的基于密度的离群点挖掘算法。

A new data exponent weighted fuzzy clustering approach is proposed by introducing a set of exponent weighting factors and influence exponent,the new approach makes it possible to treat the data points discriminatively.The new approach is combined with the existing Gustafson-Kessel（G-K） algorithm and a new algorithm,DWG-K is presented.Numerical experiments show that the DWG-K is better than G-K in improving the quality of clustering,and in the outliers mining,DWG-K detects the outliers with the global view and the physical meaning of outliers is clearer,and moreover,the computational efficiency is significantly higher than the current widely used density-based method.