The Application Of Multivariate Meta-analysis Model In Big Data Analysis

In the 21 st century,with the rapid development of Internet,especially the mobile Internet,data volume presents explosive increasing in all walks of life around the world.We have entered a new era of data technology.It is not very easy to analyze as the features of big data are large number,complex form,and unstructured.When the amount of data is large,due to the storage capacity of the computer or the limitation of software storage on variables,the data cannot be processed and analyzed at the same time,and can only be processed by blocking.In addition,the dataset sometimes needs to be stacked storage for some reasons,the researchers can not get the full amount of raw data,and also can only be processed by blocking.Usually,people tend to obtain the results of analysis of the entire data by averaging the data analysis results of the block or segmentation data.However,such block or segmentation data have heterogeneity,how to scientifically and reasonably analyze the stacked data and summarize the conclusions is worth studying.Bases on the consideration of the above issues,this paper proposes a big data analysis framework uses multivariate metaanalysis methods to help more researchers who do not have big data analysis skills to analyze data quickly and easily and to solve the problem of effectively summarizing the analysis results of the block data.Multivariate meta-analysis is a statistical method which can combines the results of different studies for one problem.This paper builds the framework of big data analysis based on multivariate meta-analysis method: combing the results of big data block analysis by multivariate meta-analysis method.Also,a large number of simulation experiments should be investigated for the effectiveness of the method.The results of simulation experiments show that: when establishing multivariate regression model,if bias is used as evaluation criterion,the effect of multivariate meta-analysis model is better than that of simple average method in 6 of the 8 settings,and when Mahalanobis distance and std are used as evaluation criteria,the effect of multivariate meta-analysis model is better in all of the 8 settings;when establishing logistic regression and increasing inter-group heterogeneity,the effect of multivariate metaanalysis model is better in all of the three evaluation criteria.That is to say,in most cases,the performance of multivariate meta-analysis is better than that of simple averaging method,and better estimation of model parameters can be obtained.Finally,the big data analysis framework based on multivariate meta analysis is applied to two real datasets.And the results show the analysis effect of multivariate meta-analysis model in the analysis of the situation of randomly spliting datasets and spliting the datasets by a feature.