Research Of Data Modeling And Algorithm Based On The Privacy Protection

With the wide application of big data in industrial production and scientific research,the privacy protection of raw data has become a very important problem.Aiming at this problem,the hierarchical structure model is designed,and the hierarchical structure data which only contains the calculation result is defined based on hierarchical structure model.Meanwhile,the multivariate linear and nonlinear regression analysis algorithms with hierarchical structure data are proposed based on the traditional regression analysis.Using the data set from the practical problem,the new algorithm is verified that the regression analysis algorithm with hierarchical structure data can effectively guarantee the privacy of the raw data and accurately calculate the partial regression coefficients of each part in hierarchical structure model.The new algorithm provides feasible solution for the analysis of big data.The main work is as follows:(1)Based on hierarchical structure data,the algorithm of partial regression coefficient for multiple linear regression analysis is proposed.During the multiple linear regression analysis with hierarchical structure data,the total partial regression coefficient is calculated by each partial regression coefficient at the lower part of the structure and the hierarchical matrix between the lower part and upper part of the structure.Based on the data set form statistical yearbook and random data,experiments validate that the new algorithm is equivalent to the traditional method of multiple linear regression.The experiments show that the new algorithm can calculate the results simultaneously.Meanwhile,the hierarchical structure data algorithm can avoid the involvement of raw data in transmission and calculation of model.From this,the new algorithm can effectively solve the problem of privacy and data protection.(2)For the more complex nonlinear model,a linearized nonlinear regression analysis algorithm is proposed based on the multivariate linear regression analysis algorithm with hierarchical structure data.Taking the Cobb-Douglas production function as an example,the calculation method of the nonlinear partial regression coefficient is given based on the linearized partial regression coefficient model.In addition,the total partial regression coefficient of nonlinear regression can be calculated by the relationship between the partial-total models with hierarchical structure.Due to statistical yearbook data,the experiments validate that the calculation result based on the total partial regression coefficient of nonlinear regression model and the calculation result based on the raw data only have accuracy error.What's more,the use of hierarchical structure data can effectively reduce the possibility of revealing raw data in the new model.(3)In order to satisfy the significant index of the hierarchical structure model with privacy data,the significant problem of the bottom layer regression algorithm is further studied.Meanwhile,this study can ensure the significant of each layer partial regression coefficients in the multiple regression model with hierarchical structure data.Due to the big data in bottom layer model and the significance of the upper layer model is guaranteed by the bottom layer model,so the statistical significance of bottom layer must overcome the problem that the increased possibility of type I error.The results of big data which come from open and mean-ing gene database validate that Multiple Hypothesis Test based on FWER can reduce the possibility of type I error.As a consequence,it is particularly important for privacy protection model to do Multiple Hypothesis Test based on hierarchical structure model.