传统K-means及其众多改进算法缺乏显式处理异常样本的能力,导致其聚类性能容易受到异常样本的影响.针对此问题,提出一种融合异常检测与区域分割的高效K-means聚类算法.首先,通过构建统一聚类模型,形成异常检测与聚类之间的交互协同,以提高聚类性能.其次,利用近邻簇搜索技术对各类簇进行自适应的区域分割,以减少冗余计算,提高算法执行效率.最后,为验证所提方法的有效性,在多个合成数据集和真实数据集上分别进行测试.实验结果表明:所提算法聚类性能和执行效率优于其他算法;在添加 10%异常样本的Wine数据集上准确度可达 0.911.
In the traditional K-means and many improved algorithms,the inability to explicitly handle outliers,re-sulted in their poor clustering performance.To solve this problem,in this paper,an efficient K-means with region segment and outlier detection was proposed.Firstly,to obtain better clustering results,an unified clustering model to form an interactive collaboration between outlier detection and clustering was constructed.Secondly,to improve algorithm efficiency,clusters were adaptively segmented through near neighbor clusters search to reduce redundant calculations.Finally,on synthetic datasets and real datasets were tested to verify the effectiveness of the proposed method.The experimental results showed that EK-means algorithm outperformed other algorithms in terms of cluste-ring performance and execution efficiency.The ACC could reach 0.911 in the Wine dataset.
尹宏伟;杭雨晴;胡文军
湖州师范学院 信息工程学院,浙江 湖州 313000||湖州师范学院 浙江省现代农业资源智慧管理与应用研究重点实验室,浙江 湖州 313000
计算机与自动化
聚类;K-means;异常检测;区域分割;近邻簇搜索;自适应
clustering;K-means;outlier detection;region segment;near neighbor clusters search;adaption
《郑州大学学报（工学版）》 2024 (003)
80-88 / 9
国家自然科学基金资助项目(62206094);湖州市公益性应用研究项目(2021GZ05);江苏省网络空间安全工程实验室开放课题(SDGC2237)
10.13705/j.issn.1671-6833.2024.03.010
评论