因素空间理论下的因果概率推理分类算法研究  

A causal probabilistic inference classification algorithm based on factor space theory

作　　者：曾繁慧[1,2] 胡光闪 孙慧 汪培庄 ZENG Fanhui;HU Guangshan;SUN Hui;WANG Peizhuang(College of Science,Liaoning Technical University,Fuxin 123000,China;Institute of Intelligence Engineering and Mathematics,Liaoning Technical University,Fuxin 123000,China)

机构地区：[1]辽宁工程技术大学理学院,辽宁阜新123000 [2]辽宁工程技术大学智能工程与数学研究院,辽宁阜新123000

出　　处：《智能系统学报》2024年第4期1042-1051,共10页CAAI Transactions on Intelligent Systems

基　　金：辽宁省教育厅资助项目(JYTQN2023210,LJKZZ20220047);阜新市社会科学课题(2023Fsllx154,2023Fsllx017)

摘　　要：机器学习方法与因果推理结合能极大地提升方法性能。为探究因果概率正逆向推理的分类效果,基于因素空间理论下的因素概率论,利用条件概率,研究正向因素概率推理原理及模型并提出正向因果概率推理分类法(forward causal probabilistic inference classification algorithm,FCPIC)和简化条件的可取度分类法;研究逆向因素概率推理原理及模型并结合贝叶斯网络提出逆向因果概率推理分类法(reverse causal probabilistic inference classification algorithm,RCPIC)。将3个分类算法与KNN(K-Nearest neighbor)和SVM(support vector machine)算法进行实例对比验证,研究结果表明:FCPIC算法、可取度分类算法和RCPIC算法简单有效、具有可行性和实用性,且可取度分类法和RCPIC算法性能优于SVM和KNN算法,FCPIC算法对实际数据预测中必要类有查全需求的情况更优。研究结论丰富了因素空间的理论研究和应用价值。The integration of machine learning techniques with causal reasoning can significantly enhance method performance.To investigate the classification effect of positive and reverse causal probability inferences,we rely our study on factor probability theory under factor space theory.Using conditional probability,we examined the principles and model of positive-factor probabilistic reasoning.This led to the proposal of the forward causal probabilistic inference classification algorithm(FCPIC)and a desirability classification method of simplified conditions.We also explored the principles and model of inverse factor probabilistic inference,which resulted in the proposal of the reverse causal probabilistic inference classification algorithm(RCPIC)along with a Bayesian network.The three classification algorithms were compared with the K-nearest neighbor(KNN)and support vector machine(SVM)algorithms.The results demonstrate that the FCPIC algorithm,the desirability classification algorithm,and the RCPIC algorithm are simple,effective,feasible,and practical.The performance of the desirability classification method and RCPIC algorithm surpasses those of both SVM and KNN.Additionally,the FCPIC algorithm is better when dealing with cases where the necessary classes in actual data prediction have full demand.These research findings contribute to the theoretical research and application value of factor space.

关 键 词：因素空间 因果概率推理分类法 可取度分类法 贝叶斯网络 因素概率论 条件概率 因果关系 人工智能 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]