A Bayesian network model is a popular formalism for data mining due to its intuitive interpretation. This chapter presents a semantic genetic algorithm (SGA) to learn the best Bayesian network structure from a database. SGA builds on recent advances in the field and focuses on the generation of initial population, crossover, and mutation operators. In SGA, we introduce semantic crossover and mutation operators to aid in obtaining accurate solutions. The crossover and mutation operators incorporate the semantic of Bayesian network structures to learn the structure with very minimal errors. SGA has been proven to discover Bayesian networks with greater accuracy than existing classical genetic algorithms. We present empirical results to prove the accuracy of SGA in predicting the Bayesian network structures.
Benchmarking dynamic Bayesian network structure learning algorithms
