Shape-Embedded-Histograms for Visual Data Mining

Abstract

Scatterplots are widely used in exploratory data analysis and class visualization. The advantages of scatterplots are that they are easy to understand and allow the user to draw conclusions about the attributes which span the projection screen. Unfortunately, scatterplots have the overplotting problem which is especially critical when high-dimensional data are mapped to low-dimensional visualizations. Overplotting makes it hard to detect the structure in the data, such as dependencies or areas of high density. In this paper we show that by extending the concept of Pixel Validity (1) the problem of overplotting or occlusion can be avoided and (2) the user has the possibility to see information about an additional third variable. In our extension of the Pixel Validity concept, we summarize the data which are projected onto a given region by generating a histogram over the required attribute. This is then embedded in the visualization by a pixel-based technique.