Uncertainty Oriented Visual Multidimensional Data Analysis

Multidimensional data analysis is an important and challenging problem in many areas,such as academical research and industrial applications.With the development of technology,the scale and complexity of multidimensional data increase.The research of multidimensional data visualization and visual analytics have also attracted more atten-tion.In order to better assist users in understanding data and making reasonable decisions,uncertainty visualization has also become an important and hot topic in visualization re-search.How to take advantage of uncertainty to help users in multidimensional data exploration and analysis is still a challenging problem.Moreover,how to use multidimen-sional data analysis methods to help evaluate data uncertainty is an urgent and meaningful research topic.This paper mainly focuses on several related problems on uncertainty oriented multi-dimensional data analysis and propose effective solutions for them.The main contributions of this paper include:· For level of detail multidimensional data and dimensional correlation analysis,an uncertainty based visual data exploration method is proposed.Based on scatterplot,the method takes advantage of super pixel based multilabel optimization to generate clustering results with high data abstraction quality.A set of glyph are designed to visualize the data uncertainty of clusters and guide users in level of detail data exploration.The method can support interactive data exploration even for large scale data.· For multidimensional data classification,an uncertainty visualization based active learning method is proposed.It helps users with sample selections in active learning.Based on the iso-contour based uncertainty visualization on the scatterplot,users can select samples by taking sample uncertainty,diversity and density simultaneously into consideration.This can ensure the effectiveness of samples and improve the learning efficiency of the classification model.Qualitative and quantitative evaluations are used to prove the usability of the method.· To evaluate the uncertainty of multidimensional weather forecast,a visual voting framework is proposed and a visual analytics system is designed and developed.The framework uses the analog method,which has already been proved to be useful in the domain research,to evaluate the uncertainty of precipitation forecast(the pre-cipitation probability).Majority voting method is used to tackle the various analog similarity measurement problem and the coordination among measurements and views are used to handle the threshold problems in these measurements.Thereafter,the uncertainty of precipitation forecast can be better evaluated,compared with ex-isting analog methods with default parameters.Moreover,case studies and domain expert feedback are used to prove the usability of the method and the system.