Research And Improvement Of Naive Bayesian Classifier

Classification is an effective method for predictive modeling in data mining.By constructing a target function(also called a classifier or classification model),the model can map the data in the database to one of the given categories.In the real life it has a wide range of applications,including fraud detection,target marketing,risk assess-ment and medical diagnosis.There are many methods to classify,among them,Naive Bayesian classifier has became one of the hot research topics of the current classification algorithms with such characteristics:its unique uncertain knowledge expression form,solid statistical theory foundation,integrated incremental learning of prior knowledge as well as simple and efficient.According to the research,we found that the Naive Bayesian classification model based on conditional independence assumption can be compared with the decision tree and the neural network classification model,what's more,sometimes it can even perform better in some fields.However,in most cases,the conditional independence assumption is contrary to the reality,that will significantly reduce the classification accuracy.Based on this fact,we might raise the question that whether to relax the conditional independence assumption of the Naive Bayesian classifier can improve the classification performance.In this paper,we use Naive Bayesian classification model as the basic object,also,the four kinds of improved models are introduced.At the same time,the improved method based on singular value decomposition and principal component analysis is proposed for the independence assumption in this paper.Finally,we combine singular value decomposition and principal component analysis with Naive Bayesian model to improve the effect.On the basis of the above analysis,taking into account that the different attributes have different effects on the classification results,the paper puts forward another improved method-weighted Naive Bayesian classification model based on singular value decomposition and principal componen-t analysis.Finally,the simulation results show the effectiveness of the improved method.