Online Markov Blanket Learning with Group Structure

摘要

Learning the Markov blanket (MB) of a given variable has received increasing attention in recent years because the MB of a variable predicts its local causal relationship with other variables. Online MB Learning can learn MB for a given variable on the fly. However, in some application scenarios, such as image analysis and spam filtering, features may arrive by groups. Existing online MB learning algorithms evaluate features individually, ignoring group structure. Motivated by this, we formulate the group MB learning with streaming features problem, and propose an Online MB learning with Group Structure algorithm, OMBGS, to identify the MB of a class variable within any feature group and under current feature space on the fly. Extensive experiments on benchmark Bayesian network datasets demonstrate that the proposed algorithm outperforms the state-of-the-art standard and online MB learning algorithms.

关键词