Incremental learning procedure of Bayesian network structure based on improved hill-climbing algorithms
Wan Men · Journal of Lanzhou University of Technology · 2013
Because of possible greater difference between the constructed Bayesian network and domain environment as well as the varying character of the dynamics of the domains,it is very necessary to improve the performance and optimize the structure of Bayesian network as new data is observed.The traditional hill-climbing algorithm was investigated,and the hill-climbing algorithm proposed by Gamez was improved,where the forbiddenness list and cycle forbiddenness list were introduced when the node was to be deleted,so that the search of unnecessary redundant was avoided,the efficiency of searching was improved,and an update method of forbiddenness list was given.Experimental results showed that this improved algorithm was effective.