| Currently,we are facing massive and diverse data.These data come from different sources,exist in different formats and forms,and many analyses are based on a single data source.In research,using a single data source for analysis may result in incomplete relevant covariates,and therefore may lead to unsatisfactory research results.Data heterogeneity needs to be considered during data integration.Ignoring this heterogeneity in data analysis can lead to biased estimates and misleading inferences.The data fusion considered in this article is based on two aspects:(1)Fusion variables.For the information collected by multi-source sensors,three fusion methods are used:primary fusion,intermediate fusion,and advanced fusion.This article provides a detailed introduction to these three fusion methods.Using the traditional advanced fusion method,there will be the same number of votes in the voting decision,so this paper introduces the fuzzy operator,combines the posterior probability and the fuzzy operator to make the voting decision,and eliminates the situation of the same number of votes.In order to optimize the classification effect,the results of a single sensor and three fusion methods are fused.(2)Fusion samples.The data of different groups are fused.In order to solve the problem of heterogeneity during fusion,this paper uses a regularization fusion method,introduces order statistics in the penalty term to minimize the difference of coefficients between groups,and uses EBIC to select parameters.The results show that in some categories,the classification effect using data fusion is better than using a single data source,and the discrimination effect using a combination of posterior probability and fuzzy operators is better than traditional advanced fusion methods.Compared to traditional models,fusion lasso method provides richer explanations and visualization,and it allows for the identification of homogeneous parameter groups between studies without using hypothesis testing methods in regression analysis. |