基于图结构的概念漂移检测  

Concept drift detection based on graph structure

作　　者：周彦冰 马士伦 文益民[1] ZHOU Yanbing;MA Shilun;WEN Yimin(Guangxi Key Laboratory of Image and Graphic Intelligent Processing,Guilin University of Electronic Technology,Guilin 541004,Guangxi,China)

机构地区：[1]广西图像图形与智能处理重点实验室(桂林电子科技大学),广西桂林541004

出　　处：《山东大学学报(工学版)》2025年第2期88-96,共9页Journal of Shandong University(Engineering Science)

基　　金：国家自然科学基金资助项目(62366011);广西重点研发计划资助项目(桂科AB21220023);广西图像图形与智能处理重点实验室资助项目(GIIP2306)。

摘　　要：为了解决传统的概念漂移检测方法,仅依赖错误率进行漂移检测不可靠的问题,提出一种基于图结构的概念漂移检测方法。该方法使用k关联最优图表示当前数据分布,定义样本的漂移率表示分类器与当前数据分布的不一致性,利用漂移率形成比特流,使用概念漂移检测器在比特流上检测概念漂移。通过与传统的使用错误率的概念漂移检测方法的对比和分析,结果表明在人工数据集上基分类器的准确率提高1%~5%,在真实数据集上提高1%~2%。所提出的方法有效提高概念漂移检测的准确性,帮助基分类器更好适应概念漂移。In order to solve the problem that the traditional concept drift detection method only relied on the error rate for drift detection was not reliable enough,a concept drift detection method based on graph structure was proposed.In this method,the kassociated optimal graph was used to represent the current data distribution,and the drift rate of the sample was defined to represent the inconsistency between the classifier and the current data distribution.The drift rate was used to form a bit stream,and the concept drift detector was used to detect the concept drift on the bit stream.Compared with the traditional concept drift detection method using error rate,the results showed that the accuracy of the base classifier was improved by 1%-5%on artificial datasets and 1%-2%on real-world datasets.The proposed method could effectively improve the accuracy of concept drift detection and help base classifiers better adapt to concept drift.

关 键 词：数据挖掘 数据流 概念漂移 图结构 k关联最优图 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]