Mapping Mental Models of Uncertainty to Parallel Coordinates by Probabilistic Brushing

Borrelli, Gabriel; Ittermann, Till; Linsen, Lars

Abstract

Through training and gathered experience, domain experts attain a mental model of the uncertainties inherent in the visual analytics processes for their respective domain. For an accurate data analysis and trustworthiness of the analysis results, it is essential to include this knowledge and consider this model of uncertainty during the analytical process. For multi-dimensional data analysis, Parallel Coordinates are a widely used approach due to their linear scalability with the number of dimensions and bijective (i.e., loss-less) data transformation. However, selections in Parallel Coordinates are typically achieved by a binary brushing operation on the axes, which does not allow the users to map their mental model of uncertainties to their selection. We, therefore, propose Probabilistic Parallel Coordinates as a natural extension of the classical Parallel Coordinates approach that integrates probabilistic brushing on the axes. It supports the interactive modeling of a probability distribution for each parallel coordinate. The selections on multiple axes are combined accordingly. An efficient rendering on a compute shader facilitates interactive frame rates. We evaluated our open-source tool with practitioners and compared it to classical Parallel Coordinates on multiple regression and uncertain selection tasks in user studies.