Towards Quantitative Visual Analytics with Structured Brushing and Linked Statistics

• Published in: Computer graphics forum Vol. 35; no. 3; pp. 251 - 260
• Main Authors: Radoš, S., Splechtna, R., Matković, K., Đuras, M., Gröller, E., Hauser, H.
• Format: Journal Article
• Language: English
• Published: Oxford Blackwell Publishing Ltd 01.06.2016
• Subjects: Analysis; Analytics; Brushes; Brushing; Categories and Subject Descriptors (according to ACM CCS); Computer graphics; Design analysis; I.3.6 [Computer Graphics]: Methodology and Techniques-Interaction techniques; Medical diagnosis; Quantitative analysis; Statistics; Studies; Visual; Visualization
• DOI: 10.1111/cgf.12901
• ISSN: 0167-7055

Until now a lot of visual analytics predominantly delivers qualitative results—based, for example, on a continuous color map or a detailed spatial encoding. Important target applications, however, such as medical diagnosis and decision making, clearly benefit from quantitative analysis results. In this paper we propose several specific extensions to the well‐established concept of linking&brushing in order to make the analysis results more quantitative. We structure the brushing space in order to improve the reproducibility of the brushing operation, e.g., by introducing the percentile grid. We also enhance the linked visualization with overlaid descriptive statistics to enable a more quantitative reading of the resulting focus+context visualization. Additionally, we introduce two novel brushing techniques: the percentile brush and the Mahalanobis brush. Both use the underlying data to support statistically meaningful interactions with the data. We illustrate the use of the new techniques in the context of two case studies, one based on meteorological data and the other one focused on data from the automotive industry where we evaluate a shaft design in the context of mechanical power transmission in cars.