基于粒球邻域粗糙集的三支高斯混合聚类

打开文本图片集

中图分类号：TP391 文献标志码：A 文章编号：1671-6841（2025）06-0016-08

Abstract： In order to solve the problem of redundant information in afecting the clustering effect of three-way Gaussan mixture models in high-dimensional datasets，the theory of granular ballneighborhood rough sets was integrated into the model，and a three-way Gaussian mixture clustering model based on granular ball neighborhood rough sets was proposed. Firstly， k -means clustering was used to generate a set of granular balls that meet the purity requirements，and atribute reduction was performed with the invariant constraint of the positive region produced by the granular balls to extract key atributes. Secondly， the three-way Gaussan mixture model was used to cluster the reduced data，dividing the objects into the core region or the boundary region of the clusters. Comparative experimental results on 7 UCI public datasets demonstrated that the proposed model not only inherited the superior clustering performance of the three-way Gaussian mixture model with higher accuracy，silhouette coeffcient，and lower Davies-Bouldin index，but also provided a more accurate depiction of the cluster boundaries. Furthermore，as a result of reducing atributes in high-dimensional space，the proposed model achieved lower time complexity.

Key Words：high-dimensional data；three-way Gaussan mixture model; clustering；granular ball neighborhood rough set；positive region；attribute reduction

0 引言

高斯混合模型（Gaussian mixture model，GMM）的实质是通过多个高斯函数的加权平均和来刻画样本数据的分布。（剩余12179字）