基于因果发现的业务流程概念漂移根因分析

打开文本图片集

中图分类号：TP391 文献标志码：A 文章编号：1001-3695（2025）12-012-3619-09

doi：10.19734/j. issn.1001-3695.2025.06.0171

Causal discovery-based root cause analysis for business process concept drift

Shang Xinyu’，Lu ，Fang Xianwen1，2 （1.ScholofMathematicsandBigData，AnhuUniersityfScienceandTchnolgy，HuainanAnhu22ol，China;2.AnhiEin Laboratory ofCoal Mine Safety forBig Data Analysis and Early Warning Technology，HuainanAnhui 232Oo1，China）

Abstract：Business process models evolve over time，causing models built from historical eventlogs to loseaccuracy.Detectingconceptdriftoptimizes processmodels toadapttoenvironmentalchanges，andanalyzingdriftcausesprovidesabasis for optimization.Existing concept drift detection techniques mainlyrelyonchanges inactivityrelationships withincontrolflows. These techniques overlok thevarying influenceofactivityrelationshipsondrift，made ithard toexplainrootcauses.Toaddressthis，this paper proposed CADDAR（causaldriftdetectionandrationalization），atechnique basedonchanges in feature influence.Specificall，the methodusedactivity pairs incontrolflows and processdurationasfeatures andoutcomesforcausal discovery.Itexaminedcausalcoeficients，treatedthemastheinfluenceoffeaturesondrift.Thenthemethodselectedactivity pairs with significant influenceascausalfeatures，andusedthechanges intheinfluenceofthesecausalfeatures to detectdrift， whileasliding window pinpointeddriftlocations.Finally，threetypesofchanges incausalfeatureinfluence—changes incausalelationshipsand theirstrength，servedasrootcausesofdrift.ExperimentsshowthatCADDARoutperforms existing techniques.Case studies further demonstrate that this method effctively explains the root causes ofconcept drift.

Keywords：：business processes；eventlog；processmining；causal discovery；influeneofcharacteristics；conceptdrift；drift explanation

0 引言

测和分析。（剩余20601字）