Local Bayesian Network Structure Learning for High-Dimensional Data

To address the challenge of achieving higher learning accuracy and efficiency in local Bayesian network structure learning for high-dimensional data, we introduce a new algorithm combines feature selection with the Meek rules to construct local Bayesian network structures, known as FSCLBN. Our experimental results show that the average F1 scores of FSCLBN and the other three algorithms (PCD_by_PCD, CMB, MY_by_MB) on all data sets are 0.56, 0.37, 0.44 and 0.21 respectively. Therefore, the FSCLBN algorithm outperforms traditional local Bayesian network structure learning methods when dealing with high-dimensional data sets.