Visualization Enhancement of Parallel Coordinates Plot Based on Fisher Score and Laplacian Score

Abstract

Over the past decades, high-dimensional data visualization analysis has always been a hot topic in the field of data science. PCP (Parallel Coordinate Plot) is a very commonly utilized tool in the field of data analysis. To be specific, each feature of the dataset can be illustrated in a Cartesian Coordinate System. To complete the recording on one data from a dataset onto the chart, one needs to find the numerical value of each feature belonging to one data on each feature axis and connect those points on each feature axis together. However, when using PCP to deal with and analyze a large amount of data and features, overlapping and crossing between segments would strongly affect the visualization performance of the chart and therefore increase the difficulty of data analysis. To address such issue, this paper presents a visualization enhancement method that can reorder feature axes on the plot and remove unnecessary feature axes automatically. To reorder and remove feature axes automatically in PCPs, we employed the Fisher score and Laplacian score to reorder features based on the corresponding weight. By comparing the visualization result of reordering for each method, features with low priority among the reordering result of both methods can be observed. After using this method on PCP, the visualization performance of PCPs considerably improved, which demonstrates that the methods based on feature selection are beneficial to optimize the PCP performance.