A Research On Correlation Analysis Of Dimensional Uncertainty

With the progress of data analysis technology,uncertainty and correlation,as important elements of data analysis,has been widely used.Correlation measures the changing relationship between dimensions,while the correlation based on uncertainty measures the changing relationship between uncertainty of dimensions.In addition,Whether the correlation based on uncertainty and the original data are the same relationship or not ? If not,what are their similarities and differences.How to locate the source of the uncertainty and analyze the data feature that create additional uncertainty by using this correlation based on uncertainty?Aiming at the above research objectives,this paper proposes an uncertainty analysis framework and a visual interaction system.The framework conducts clustering analysis using K-medoids,and obtains several groups according to the result of clustering analysis,firstly.Then it constructs the uncertainty data set through calculating the uncertainty of each group by the uncertainty quantification method.In addition,to verify the rationality of the constructed uncertainty data sets,the result will be tested to prove the rationality.Finally,the correlation analysis and comparative analysis of these two correlations will be executed to find the similarities and differences.The visual system provides scatter plots,correlation diagrams and a parallel graph to help user analyze the row data,compare the correlations and discovery the distributional features,respectively.Finally,this paper will use two public data sets for case analysis to verify the rationality and effectiveness of the framework and system.And it comes to the following conclusion.The two correlations have similarities in the dimension of high uncertainty,but there are differences in the dimension of low uncertainty.In addition,the data features that generate additional uncertainties exhibit a distribution that diverges from the less uncertain dimension to the higher dimension.