Discovering and orienting the edges connected to a target variable in a DAG via a sequential local learning approach

Given a target variable and observational data, we propose a sequential learning approach for discovering direct cause and effect variables of the target under the causal network framework. In the approach, we start from the target, sequentially find Markov blankets of variables and learn local structures over Markov blankets until we determine the causes and effects of the target variable. Without constructing a whole network over all variables, we find only a local structure around the target. The main advantage of the proposed sequential approach is that it can greatly reduce CPU time compared with whole network learning approaches for finding the causes and effects of a given target node in a large network. The proposed approach can be applied to predict the effects of external interventions. Theoretically we show the correctness of the approach under the assumptions of faithfulness, causal sufficiency and that independencies are correctly checked. These theoretical results can also be used for learning directed acyclic graphs with latent variables.