MultiVisD3: uma abordagem visual analytics para análise da qualidade de projeções multidimensionais utilizando múltiplas visões coordenadas

Abstract

Multidimensional visualizations are graphical representations that act in the presentation of multidimensional or multivariate datasets. However, many of these techniques exhibit poor visual scalability as dimensions increase, requiring an increasingly larger visual space to accommodate the entire dataset. One way to get around this problem is to use Multidimensional Projections techniques that perform dimensionality reduction and seek to preserve the patterns of the original data in the projected space. During this process, it is highly likely that errors and distortions will be present in the projection layouts. A challenge in researching this topic is measuring and visualizing the errors and distortions incorporated in the projection mappings. In this sense, this work developed MultiVisD3 a Visual Analytics approach to analyze the quality of Multidimensional Projections. This approach comprises a dashboard as an interactive visual panel containing multiple coordinated views to facilitate understanding and comparison of error metrics between two projections. Other visualization techniques such as Treemap and tabular view support the extraction of information about the quality of projections. Finally, the proposed approach was subjected to a user evaluation using known datasets to analyze aspects of functionality and usability. The results showed that users positively evaluated the approach proposed in this work. More than 90\% of participants correctly performed eight of the nine tasks asked for them and more than 90\% of them approved the usability aspects related to interactions and visualization arrangement.