Research On Bayesian Network Structure Learning Based On Artificial Bee Colony Algorithm

Bayesian network is also called bays net, belief network or causal probabilitynetwork. It combines graph theory and statistical model and can intuitively indicatesdependence and independence relationships between random variables. Bayesiannetwork has well-defined semantics and solid theoretical foundations. It has beensuccessfully applied to artificial intelligence, machine learning, data mining,bioinformatics, etc. Usually, constructing a Bayesian network only depend on thedomain expert is difficult, even impossible. Therefore, the central issue and difficultpoint is how to learn Bayesian networks in the data through the efficient methods andalgorithm. The learning Bayesian networks mainly includes: structural and parameterlearning, parameter can be curtained through networks structure and data sets, sostructural learning is the core of learning Bayesian networks.Firstly, make a comprehensive introduction to relevant theories of Bayesiannetwork, and get deep research on the two basic methods of Bayesian networkstructure learning.Secondly, the basic theory and definition of the meta-heuristic-artificial beecolony optimization algorithm are introduced in detail. Taking into account the basicartificial bee colony algorithm converges slowly and prematurely, an improvedartificial bee colony algorithm based on local search is proposed. The method makesfull use of the stochastic dynamic local search to optimize the current best solution tospeed up the convergence rate. In order to maintain the population diversity and avoidpremature convergence, the selection probability based on ranking is used instead ofdepending on fitness directly.Finally, Bayesian network structure learning method based on artificial bee colonyalgorithm is proposed. The algorithm uses MWST algorithm to generate the treestructure of the network, and form initial population through three operating operator.According to the characteristics of Bayesian network structure, the updated strategy ofnectar is designed. Simulation results show that the algorithm is valid.