Semi-simple Bayesian Classifier Research

With the coming of big data era,we have got a new understanding of the data.We can do more and more through the accumulation of data,which is especially true in the importance of Data mining.As an important direction of data mining,bayesian classification is achieving a high-speed development as simple and efficient in the aspect of the data processing.We have a common pursuit in achieving the data we need from huge amounts of data fast and efficient.At the moment,Bayesian classification has obtained widespread application in lots of fields,and it has become a research hotspot in artificial intelligence.With the successful application in data mining,naive bayesian classifier inspires us to do the research of classification accuracy from the aspect of exploration to reduce attribute independence assumption.Research about this kind of algorithm is roughly divided into the following categories:First,split the attributes into subsets in the traditional method of classification,and then use naive bayesian classifier again;Second,allow the interdependencies between attributes to improve the naive bayesian classifier;Third,the classification algorithm of selecting the attributes;Forth,use probability estimation to calibrate naive bayesian classifier,and introduce hidden variables to improve naive bayesian classifier.This article analyzed the improved algorithm,it put forward two improved classifiers:the improved bayesian classifier based on the greedy selection algorithm and Fully Bayesian classifier.Greedy selection algorithm is a common used algorithm of finding the optimal solution,in this paper,we use it to find optimal group of attributes in improving naive Bayesian classifier.Fully Bayesian classifier is based on the bayesian theory,it set the prior distribution of unknown parameters,and consider the uncertainty of other parameters in the process of inference to improve classifier.On the basis of the two methods above,we use different principles to divide property group and set the parameters in the process of classification,and we have achieved good results in the process of a large number of data sets classification prediction.In this paper,we applied the classification algorithm to many aspects,for example:the national macroeconomic data classification,the classification predition of the enterprise management risk and financial risk.After the classification experiment and application data,through classification experiment and application data,we get the result that these two aspects of improvement is reasonable and effective.