Related Research On Bayesian Network Classifier

Classification is the core issue in the field of pattern recognition and machine learning research. The main research content of classification is how to find the relationship between class variable and attributes variables from the observed data and use the relationship to build a classifier to predict unknown data. Among all the classifiers construction method, Bayesian network classifier attracted more attention due to its solid theoretical basis and high classification accuracy rate.In order to improve the classification performance of Bayesian network classifiers, this paper made the following two innovations.(1) For the performance of traditional KDB(k-dependence Bayesian network classifier) couldn’t enhance just using Boosting algorithm, a KDB integrated classifier learning algorithm(BLCKBD) is proposed through relaxing the dependent conditions of traditional KDB. First it relaxed the condition of traditional KDB(selecting the strongest dependent relationship) to get different classifiers through using different parameters. Then integrate these classifiers using Boosting algorithm to achieve an ensemble classifier with higher classification accuracy. Experimental results show that the new algorithm(BLCKBD) not only has higher classification accuracy rate than KDB(especially on the dataset with more attribute), but also has better generalization ability.(2) A multi smoothing parameter na?ve Bayesian network is proposed. It uses the mean square error(MISH) to measure the deviation between estimated density and actual density and obtain the optimal smoothing parameter for every attribute through minimizing MISH. Thus a multi smoothing parameter na?ve Bayesian classifier is constructed. The experimental results show that the new algorithm has higher classification accuracy.