Pigeon Inspired Optimization of Bayesian Network Structure Learning and a Comparative Evaluation

Abstract

Bayesian networks are useful analytical models for designing the structure of knowledge in machine learning. Probabilistic dependency relationships among the variables can be represented by Bayesian networks. One strategy of a structure learning Bayesian Networks is the score and search technique. In this paper, we present a new method for structure learning of the Bayesian network which is based on Pigeon Inspired Optimization (PIO) Algorithm. The proposed algorithm is a simple one with fast convergence rate. In nature, the navigational ability of pigeons is unbelievable and highly impressive. In accordance with the PIO search algorithm, a set of directed acyclic graphs is defined. Every graph owns a score which shows its fitness. The algorithm is iterated until it gets the best solution or a satisfactory network structure using map and compass, and landmark operator. In this work, the proposed method compared with Simulated Annealing, Bee optimization and Simulated Annealing as a hybrid algorithm, Bee optimization and Greedy search as a hybrid algorithm, and Greedy Search using BDeu score function. We also investigated the confusion matrix performances of the methods. The paper presents the results of extensive evaluations of these algorithms based on common benchmark data sets. The results indicate that the proposed algorithm has better performance than the other algorithms and produces higher scores and accuracy values.