According to the low efficiency and low accuracy of the naive Bayesian network classification model in dealing with large number of high-dimensionaldata,by combining Principal Component Analysis and K-means clustering algorithm,this paper gives an improved Naive Bayesian network calssification model.The model abandoned the premise for the relative independence between non-calss attribute variables and class attribute variables.Firstly,we use principal component analysis to reduce the dimensionality of the data set,so the algorithm can focus on the classification problem.The algotithm has also proposed aconcept called "relative fusion point" to effectively improve the performance of the algorithm.Finally,the performance of the algorithm is analyzed,and the improved algorithm is applied to the actual data set for experiment to repair the missing data of the data set,the results show that the algorithm is effective.