Return to author index

Volume: 5 Issue: 1

A novel hybrid Bayesian network structure learning algorithm based on correlated itemset mining techniques

Zahra Kebaili, Alex Aussem

Abstract:
In this paper, we propose a novel hybrid method for bayesian network structure learning that combines ideas from datamining, constraint-based and search-and-score learning. The aim of this method is to identify and to represent conjunctions of boolean factors implied in probabilistic dependence relationships, that may be ignored by constraint and scoringbased learning proposals when the pairwise dependencies are weak (e.g., noisy-XOR). The method is therefore able to identify some specific high order interactions that cause the violation of the faithfulness assumption on which are based most constraintbased methods. The algorithm operates in two steps: (1) extraction of supported and minimally correlated itemsets, and (2), construction of the structure by these itemsets. The method is illustrated on a simple but realistic benchmark plaguing the standard scoring and constraint-based algorithms.

doi:10.5019/j.ijcir.2004.166

Full Text PDF
















^ TOP