Empirical Guidance on Scatterplot and Dimension Reduction Technique Choices

Abstract

To verify cluster separation in high-dimensional data, analysts often reduce the data with a dimension reduction (DR) technique, and then visualize it with 2D Scatterplots, interactive 3D Scatterplots, or Scatterplot Matrices (SPLOMs). With the goal of providing guidance between these visual encoding choices, we conducted an empirical data study in which two human coders manually inspected a broad set of 816 scatterplots derived from 75 datasets, 4 DR techniques, and the 3 previously mentioned scatterplot techniques. Each coder scored all color-coded classes in each scatterplot in terms of their separability from other classes. We analyze the resulting quantitative data with a heatmap approach, and qualitatively discuss interesting scatterplot examples. Our findings reveal that 2D scatterplots are often 'good enough', that is, neither SPLOM nor interactive 3D adds notably more cluster separability with the chosen DR technique. If 2D is not good enough, the most promising approach is to use an alternative DR technique in 2D. Beyond that, SPLOM occasionally adds additional value, and interactive 3D rarely helps but often hurts in terms of poorer class separation and usability. We summarize these results as a workflow model and implications for design. Our results offer guidance to analysts during the DR exploration process.

Towards Perceptual Optimization of the Visual Design of Scatterplots

“…These design factors were consistent with the scagnostics (aspects of scattered points, such as outlier, shape, trend, density, and coherence) defined by Wilkinson et al [5]. A follow-up study [17] indicated that a 2D scatterplot is as good as an interactive 3D scatterplot or a scatterplot matrix in showing classes. Gleigher et al [18] explored the visual design of multiclass scatterplots for the task of identifying the mean value of a set of points.…”

Perception-Based Evaluation of Projection Methods for Multidimensional Data Visualization

“…This conclusion was also stated by Sedlmair [45], in the sense that no single dimensionality reduction technique can be considered superior to all others, and some techniques may not reveal certain structures in real world data sets. Moreover, we observe that the projections tend to perform better on document data.…”

Visual Interaction with Dimensionality Reduction: A Structured Literature Analysis

“…In-depth investigation of effectiveness, however, goes beyond the scope of this paper. Previous work has shown that assessing effectiveness of interactive DR solutions depends heavily on context factors, such as users, data, domain, and tasks at hand [50,52]. A generic comparison of the scenarios' usefulness and effectiveness is thus a non-trivial endeavor, and further work is needed.…”