ITNO-K2PC: An improved K2 algorithm with information-theory-centered node ordering for structure learning

The Bayesian Network is considered as one of the most efficient theoretical models in the uncertain reasoning and knowledge representation fields. The Bayesian Network construction consists of two main phases: Structure learning and Parameter Learning. Actually, determining the correct structure represents a major challenge that had been widely studied and still needs to be resolved. In this paper, we introduce a novel algorithm combining an improvement of score based algorithm with an effective nodes ordering method. Based on the parents searching principle of K2 algorithm, we proposed an extension of the used search-space. Besides, using graph acyclic propriety and information theory, we introduce node ordering method. Experiment simulations on the well-known networks prove that the proposed algorithm can efficiently and accurately extract the nearest topology of the original. The real application of our method exhibits its efficiency in the analysis of the phosphate laundry effluents' impact on the watershed in Gafsa area (southwestern Tunisia).CO 2020 The Authors. Production and hosting by Elsevier B.V. on behalf of King Saud University. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).