Research On Learning Local Causal Structural Of Bayesian Networks And Its Application

In the past decades, a great deal of research has focused on learning CausalBayesian Networks from observation data. Causal structure learning is an importantcausal knowledge discovery method to disclose the nature of causal interactions inthe Bayesian Networks. But learning global structure of Bayesian Networks fromobservational data is NP-hard, and we often interested in a local causal structure ofa target variable. Therefore, learning the local causal structure of a target variablebecame an important research field. Discover the causal structure is not only achallenging research work, but also has important scientific significance and highapplication value.In this dissertation, we propose a novel method to learn the local causal structureof a target variable. Learning the local causal structure of a target variable includestwo aspects: the first is to learn the local structure of target variable fromobservational data; the second is to discover the causal relationship between thevariables in the local model. In view of the two aspects, carried out in this articleare as follows:Firstly, we present a novel local causal structure learning method by integratingfeature selection into intervention (I-LCSL). Firstly, under the faithfulnessassumption, I-LCSL utilizes HITON-MB algorithm obtaining the Markov blanketof interested variable for generating a local model; then, we randomly select aintervention variable from the local model, and generate interventional data byrandomized experiments; finally, we use Dynamic programming algorithm to obtaina local causal structure of the interested variable by combining observational dataand interventional data. After running a series of comparative experiments on twostandard Bayesian Networks, we show that our method has excellent learningaccuracy.Secondly, we present a local causal structure learning method by integratingfeature selection into intervention called a local causal structural active learningbased on causal power (CSI-LCSL). CSI-LCSL integrates the dividing structureability of Markov blanket and causal discovery ability of intervention learning.Firstly, under the faithfulness assumption, CSI-LCSL utilizes HITON-MB algorithm obtaining the Markov blanket of interested variable for generating a localmodel; then, we select a intervention variable from the local model by usingnon-sys entropy, and generate interventional data by perfect experiments; finally,we use an exact method algorithm to obtain a local causal structure of the interestedvariable by combining observational data and interventional data. After running aseries of comparative experiments on two standard Bayesian Networks, we showthat our method has excellent learning accuracy.These new methods have excellent learning performance and can deal with wellthe complex networks. Extensive experiments validate the effectiveness of ourmethod against other algorithms.