Star Plot Visualization of Ultrahigh Dimensional Multivariate Data

摘要

Visualization-based analysis of multivariate data suffers from a high degree of clutter when the number of dimensions (variables) becomes too large. Here, we extend the standard star plot technique to visualize large datasets of ultrahigh number of dimensions by a) drawing overlapped star plots with one star per data item, b) shifting the origins of radial axes away from the central point to open space in the low-value scale, and c) dynamically partitioning the dimensions into groups and mapping them to different concentric circular regions to provide multilevel star plot visualization. Our test on multivariate datasets of high dimensionality suggests that the proposed extensions with appropriate interaction options can handle large number of dimensions of potential relevance to big data analytics.