Authors: Mandy Keck, Dietrich Kammer, Rainer Groh
Abstract: When working with a given high-dimensional data set, data analysts often use different machine-learning algorithms to calculate clusters and classifications. However, it is difficult to ascertain which differences occur across these algorithm versions. For a human-readable visualization, dimensionality reduction methods are used to achieve two-dimensional mappings. In previous work, we proposed to use glyphs to support analysis tasks in these two-dimensional plots. In this paper, we investigate novel glyph-based strategies that can be used for comparison tasks. We present two interface concepts that can be applied on fixed and varying position data to compare different versions of machine-learning algorithms.