基于聚类集成选择的随机森林聚类方法  

Random forest clustering method based on cluster ensemble selection

作　　者：李金玉 刘静玮 杜明晶 吴福玉 LI Jin-yu;LIU Jing-wei;DU Ming-jing;WU Fu-yu(Jiangsu Key Laboratory of Educational Intelligent Technology,School of Computer Science and Technology,Jiangsu Normal University,Xuzhou 221100,China;Institute 706,Second Academy of China Aerospace Science and Industry Corporation,Beijing 100854,China)

机构地区：[1]江苏师范大学计算机科学与技术学院江苏省高校教育智能技术重点实验室,江苏徐州221100 [2]中国航天科工集团第二研究院七〇六所,北京100854

出　　处：《计算机工程与设计》2025年第4期990-996,共7页Computer Engineering and Design

基　　金：国家自然科学基金项目(62006104);江苏师范大学研究生科研与实践创新基金项目(2024XKT2591)。

摘　　要：为解决一些决策树受到数据噪声等因素的影响,导致它们对随机森林聚类产生有限甚至负面贡献这一问题,提出一种基于聚类集成选择的随机森林聚类方法(random forest clustering method based on cluster ensemble selection,RFCCES)。将每一棵决策树视为一个基聚类器,根据基聚类器集合的稳定和不稳定性设计两种不同的聚类集成选择方法,将评估单个决策树对随机森林的增益问题,转化为基聚类器对最终的聚类集成结果的增益问题。该算法与5种对比方法在10个数据集上进行比较,实验结果验证了RFCCES的独特优势和整体有效性。To address the issue of decision trees being affected by data noise and other factors,which can limit or even negatively impact their contribution to random forest clustering,a method called random forest clustering based on cluster ensemble selection(RFCCES)was proposed.Each decision tree was treated as a base clusterer,and two distinct cluster ensemble selection methods were developed based on the stability and instability of the base clusterers.The evaluation of the gain was transformed from a single decision tree within a random forest into the evaluation of the gain from the base clusterers in relation to the final cluster ensemble result.The RFCCES was compared with five contrasting methods across ten datasets.Experimental results verify the unique advantages and overall effectiveness of RFCCES.

关 键 词：随机森林 聚类 决策树 稳定性 聚类集成 基聚类器 聚类集成选择 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]