Enhancing parallel coordinates and RadVis visualizations using single-and multi-objective optimization

Abstract

Data visualization is crucial to discover hidden patterns and relationships in high dimensional datasets; visualization is an essential branch in data analytics applied in science and engineering fields. This thesis has targeted two state-of-the-art methods from two powerful families of visualization techniques: one with dimension reduction, Radial Coordinate Visualization (RadViz), and the other without dimension reduction, for instance, Parallel Coordinates Plot (PCP). In improving these techniques, evolutionary algorithms have been utilized to determine the optimal ordering of coordinates by considering single- and multi-objectives; using this concept, a smart mutation operator has been proposed and tested comprehensively. In order to investigate the performance of visualization proposed schemes, a benchmark dataset has been proposed, and objective and subjective assessments have been conducted. This investigation shows that the optimal ordering of coordinates can influence crucially visualization results. This thesis’s findings can be utilized to enhance other largescale visualization techniques used in visual-data analytics areas.