Data Analytics Enhanced Data Visualization and Interrogation with Parallel Coordinates Plots

Abstract

Parallel coordinates plots (PCPs) suffer from curse of dimensionality when used with larger multidimensional datasets. Curse of dimentionality results in clutter which hides important visual data trends among coordinates. A number of solutions to address this problem have been proposed including filtering, aggregation, and dimension reordering. These solutions, however, have their own limitations with regard to exploring relationships and trends among the coordinates in PCPs. Correlation based coordinates reordering techniques are among the most popular and have been widely used in PCPs to reduce clutter, though based on the conducted experiments, this research has identified some of their limitations. To achieve better visualization with reduced clutter, we have proposed and evaluated dimensions reordering approach based on minimization of the number of crossing pairs. In the last step, k-means clustering is combined with reordered coordinates to highlight key trends and patterns. The conducted comparative analysis have shown that minimum crossings pairs approach performed much better than other applied techniques for coordinates reordering, and when combined with k-means clustering, resulted in better visualization with significantly reduced clutter.