一种基于残差集群的因果发现方法的研究  

A Causal Discovery Method Based on Residual Clusters

作　　者：郭子诚 刘必健[1] 张洁[1] 吴荣珍[1] 叶惠仙[1] 邓丽萍[1] GUO Zicheng;LIU Bijian;ZHANG Jie;WU Rongzhen;YE Huixian;DENG Liping(Department of Information Engineering,Fujian Vocational College of Agriculture,Fuzhou Fujian 350000,China)

机构地区：[1]福建农业职业技术学院信息工程学院,福建福州350000

出　　处：《辽宁科技学院学报》2024年第6期34-38,共5页Journal of Liaoning Institute of Science and Technology

基　　金：福建农业职业技术学院2024年校级科技研究项目“大数据决策树算法在智慧农业精准控制系统的应用与实现”(2024JS024);福建农业职业技术学院2024“揭榜挂帅”项目“基于大数据驱动的人工智能实验教学辅助模型及应用研究”(2024JS004)。

摘　　要：基于回归误差的因果推理方法易受到小干扰的影响而导致结果发生改变,缺乏鲁棒性。针对此问题文章提出一种基于残差集群的因果发现方法。首先,利用样本量估计方法和分层抽样方法构建出多个样本量为m的数据子集Di;然后,对Di建立均衡关系模型并计算两个方向上模型的绝对值残差集,形成残差集群;最后,设计一种分析正反方向上残差集群间差异的方法,将残差集群弱小的方向推理为因果方向。实验结果表明新方法能够有效地推理出正确的因果方向,并在鲁棒性和准确率的综合性能上优于对比方法。Causal reasoning methods based on regression errors are prone to changes in results due to small disturbances and thus lack robustness.To address this issue,this paper proposes a causal discovery method based on residual clusters.First,multiple data subsets Di with a sample size of m are constructed using sample size estimation methods and stratified sampling methods.Then,an equilibrium relationship model is established for Di,and the absolute value residual sets of the models in two directions are calculated to form residual clusters.Finally,a method is designed to analyze the differences between residual clusters in posi⁃tive and negative directions,and the direction with weaker residual clusters is inferred as the causal direction.Experimental re⁃sults show that the new method can effectively infer the correct causal direction and outperforms the comparison method in the com⁃prehensive performance of robustness and accuracy.

关 键 词：数据挖掘 因果发现 因果方向 残差集群 集群差异 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]