Research On Visual Analytics Of High-dimensional Multi-subject Time-varying Data Based On Student Performance

There is a lot of valuable information and knowledge hidden in the data of students’ performance in colleges and universities over the years.Effective analysis of these data is of great significance to deeply explore the characteristics of students’ academic performance and improve instructors’ teaching levels.However,student performance data involves students,courses,instructors and other analytical subjects.These subjects have high-dimensional attributes and the associations between them are complex.Moreover,some attributes also have time-varying characteristics.How to effectively mine valuable association patterns from such complex high-dimensional multi-subject time-varying data is a great challenge.Therefore,based on data analysis methods and visualization technology,this thesis proposes and constructs a visual analysis framework MAPVis(Visual Analysis of Multi-level Association Patterns)based on a multi-level association analysis process,which supports users in deeply exploring the potential attribute associations and time-varying patterns in high-dimensional multi-subject time-varying data.Firstly,in order to intuitively reveal the characteristics of multi-dimensional attribute distribution in data,MAPVis designs the ordered heatmap matrice that can support attribute keyword ranking.Secondly,so as to deeply explore the feature patterns among multiple subjects,an interaction-rich analysis view is designed for biclustering algorithms,and a curve graph is integrated into the bipartite graph to reveal the structure and the intrinsic characteristics of biclusters.Then,in order to effectively mine the time-varying patterns in data,the clustering analysis method is combined with parallel coordinate technology to realize interactive and deep mining of time-varying patterns.At the same time,aiming at the problem that parallel coordinate technology can’t effectively present dimension information and the relationship between dimensions,a novel parallel coordinate axis expansion layout is designed.Finally,for purpose of effectively mining more details of patterns from the individual level,an information measurement algorithm based on Hamming distance is proposed to measure individual information.Moreover,a number of collaborative interactive visual views are provided to support the effective exploration of individual patterns.MAPVis also realizes the linkage and interaction among multiple groups of association analysis views through deploying rich multi-view interactive linkage technology,thus effectively supporting the collaborative analysis of multiple association patterns.On the basis of the above research,this thesis used the real historical student achievement data to do case analysis and invited domain experts to trial and evaluate.Through the evaluation feedback of domain experts,the effectiveness and practicability of MAPVis were further confirmed.This thesis provided an effective solution for users to explore complex associations in high-dimensional multi-subject time-varying data.