基于贝叶斯网络的预测模型研究-概率论与数理统计专业论文.docxVIP

Abstract Title: Research of Prediction Model based on Bayesian Network Major: Probability and Statistics Name: Rong KANG Signature:_S晗 kANG Supervisor: Prof.Desheng ZHANG Abstract Signature: 陇1ht4zJ叶片叶 Based on the analysis of existing methods of the CPI for巳cast methods，this dissertation mainly studies the Bayesian nclwork n，odel based on principal component analysis 得ming at the limitation of existing prediction method. Firstly，analysis the factors of th巳 CPI in order to reduce the prediction complexity，and to reduction the dimensiorr of influence factors use the PCA method. Then BN model is built to fit and forecast CPI based on dimension reduction. The empirical results show that: the method improv巳s th巳 accuracy of calculation results . The main research contents are as follows: 1.The paper analyses the description statistical characteristics of CPI sample data，and then detailed analysis the various factors affecting the CPI. Finally，according to the actual situation in China，the paper selects eight factors as the main index. The diss巳rtation introduc巳s the basic theory of Bay巳sian networks，and establishes a network of the sample data use the K2 algorithm. Based on the network to establish the model， and then fitting and for巳casting use this model. The empirical results show that: the predicti?n effect of Bayesian network model is not very good，and needs further improv巳m巳nt. The PCA method is used to data pre-processing through SPSS and Matlab software. A large number of original information is retained. The affected factors of the CPI are dimension reduction ，so that reducing the input variables and facilitate modeling. First of all，to improve K2 algorithm in light of the problems that appeared in the prediction. Bay巳sian network model is established bas巳d on the improved algorithrn. Then combining with the PCA method ，Bayesian network model is established based on the PCA . This model eliminates information overlap largely，and it is also used to fit and