An Improved IAMB Algorithm for Markov Blanket Discovery

Finding an efficient way to discover Markov blanket is one of the core issues in data mining. This paper first discusses the problems existed in IAMB algorithm which is a typical algorithm for discovering the Markov blanket of a target variable from the training data, and then proposes an improved algorithm λ-IAMB based on the improving approach which contains two aspects: code optimization and the improving strategy for conditional independence testing. Experimental results show that λIAMB algorithm performs better than IAMB by finding Markov blanket of variables in typical Bayesian network and by testing the performance of them as feature selection method on some well-known real world datasets.