A Multi-dimensional Data Visualization Applying a Scatterplot Packing Technique

Zheng Yunzhu, Haruka Suematsu, Takayuki Itoh, Ryohei Fujimaki, Satoshi Morinaga, Yoshinobu Kawahara
Scatterplot is one of the most popular techniques for multi-dimensional data visualization. However, it is not always easy to effectively represent the multi-dimensional spaces by scatterplots. Various dimension reduction and scatterplot matrix (SPM) techniques have been presented to effectively represent them in a single display space. Also, various interactive techniques such as ``Rolling the Dice'' have been presented to assist the visual analytics processes. We discuss a technique to represent multi-dimensional spaces using multiple scatterplots like SPM-based techniques. Here, SPM-based techniques has a drawback that each of the scatterplots will displayed very small when all pairs of dimensions are equally displayed. On the other hand, our technique presented in this poster selectively displays a set of meaningful pairs of dimensions, instead of displaying all pairs of dimensions. It calculates scores of pairs of dimensions, and selects pre-defined number of pairs of dimensions. Then, it defines the distances or connectivity of the scatterplots generated from the selected pairs of dimensions, and calculates their ideal positions based on the distances or connectivity. Finally, it places the set of scatterplots by a rectangle packing algorithm referring their ideal positions. Consequently, the technique places similarly looking scatterplots closer in the display spaces. It is useful to understand the correlations among many dimensions.