Uncertainty-aware visual analytics

The focus of the research is representing uncertainty in visual analytics applications as well as increasing user trust in predictions. The overall objective is to research meaningful visualisations for large and heterogeneous datasets, taking into account the aspects of data quality and uncertainty. The propagation of uncertainty has been researched by the visual analytics community under the umbrella of “data wrangling”. Most of this work, however, concentrates, on preprocessing of data (i.e. data entry, data (re)formatting, data cleaning, etc.). Little work has been done to represent data quality and confidence in parallel to visualisation of outcomes of the analysis. Such visualisation is vital to support decision-making: as what constitutes an error is often context-dependent and requires human judgment of domain experts, there is a need to research how such variables can be represented in parallel to outcomes of an iterative analysis process. Visualising data and process quality is not a topic that has been explored extensively in the visualisation literature. There has been research into specific techniques for uncertainty visualisation and what uncertainty itself entails, but it tends to concentrate on pre-processing data before such data is used in applications, rather than looking at how to communicate quality indicators to support decision-making by domain experts. There is an increased interest of such users to steer data analysis, including support for a variety of stakeholders in Grapevine-powered Industries. While these users are experts in their domains, they usually have little expertise on data processing. In this research project, we will research how scalable visualisations can be designed and implemented to enable domain experts to steer the analysis process. Different abstraction levels, as well as propagation of uncertainty between these levels, will be researched to support decision-making.