Research On Bayesian Network Classifier Based On Intelligent Optimization

Bayesian network is one of the Probability Graph Model(PGM)which is based on Bayesian formula.And classifier based on Bayesian network is widely applicated in lots of fileds.However,traditional Bayesian network classifiers such as Naive Bayes classifier and Tree Augment Naive Bayes classifier have some drawbacks:First,they don’t fully consider the relationships among attributes in data set,second,they don’t change the structure of network,which means they have a fixed structure and cannot fit the data well.To alleviate this limitations,we exploit evolutionary algorithms to optimize Bayesian network classifier.The major work of this paper are described as follows:First,we propose a self-adapitve Bayesian network classifier(GO-BNC)based on genetic optimization,Self-adaptiveness means the network can change its structure automatically.Genetic algorithm is exploited here to search the relationships among the attributes and dynamically change the structure of Bayesian network to better fit the data in each iteration.Experimental results show that the proposed method leads to a high classification accuracy than some traditional classification methods on some benchmarks.Second,with the growth of nodes(attributes)in datasets,the complexity of network will increase with it.To alleviate this phenomenon,we propose Multiple Object Optimization Bayesian Network Classifier(MOO-BNC)based on GO-BNC,which can optimize classification performance and network’s complexity simultaneously.In condition of sacrificing little classification performance,the proposed method can reduce the complexity of networks.Elitist Non-dominated Sorting Genetic Algorithm(NSGA-II)is exploited here to optimize both classification performance and network complexity.Experimental results show that the proposed method has good classification accuracy and better network complexity than GO-BNC on some benchmarks.Last,we construct a face recognition system based on multiple object optimization classifier.First,collecting face data and save them in the local database.Second,preprocessing the collected data with the method of histogram equalization,then reduce the dimensions of these data and extract key features.Finally,classify the objective face with proposed algorithm.