摘要
研究目的是验证连续贝叶斯网络模型可以从断面调查数据获取因果信息。使用L1MB、TC、PCB和Two-Phase等连续贝叶斯网络结构学习算法,从美国健康和营养调查(NHANES)提供的真实断面调查数据,获取潜在的因果关系。实验结果表明这些算法能不同程度地从横断面调查数据发现相应的因果关系,适用于高斯和非高斯数据的PCB算法,以及Two-Phase算法的学习性能优于仅适用于高斯数据的L1MB算法和TC算法。结合PCB算法和Two-Phase算法进行因果分析,这样得到的因果结构才较为全面。
This study is to verify that the continuous Bayesian Network model can be used to discover causal information from cross-sectional survey data. Using the L1 MB algorithm, TC, PCB and Two-Phase algorithm, this paper analyzes causal relations in the real data from the National Health and Nutrition Examination Survey. Experimental results show that these algorithms can discover causal relations to various degrees. The PCB algorithm and Two-Phase algorithm that apply to Gaussian or non-Gaussian data outperform the L1 MB algorithm and TC algorithm that only apply to Gaussian data. Combining the PCB algorithm and Two-Phase algorithm for causal analysis, the causal structure thus obtained is more comprehensive.
出处
《计算机工程与应用》
CSCD
2014年第19期192-198,共7页
Computer Engineering and Applications
基金
国家高技术研究发展计划(863)(No.2012AA011005)
安徽省科技攻关计划科技强警专项项目(No.1001130612)
关键词
贝叶斯网络
断面调查数据
因果模型
结构学习
因果关系
医学论证
Bayesian network
cross-sectional survey data
causal model
structure learning
causal relations
medical arguments
