Universität zu Köln

Braun, Daniel ORCID: 0000-0002-8824-7184 (2025). Visual Communication of Large Value Ranges and Visual Validation of Regression Models. PhD thesis, Universität zu Köln.

PDF Dissertation_DanielBraun_veroeffentlicht.pdf Bereitstellung unter der CC-Lizenz: Creative Commons Attribution. Download (58MB)

Abstract

This dissertation investigates two underexplored challenges in visual data analysis: the effective communication of large value ranges and the visual validation of statistical (regression) models. Within the visual analytics framework, both challenges target critical parts in the data analysis process — namely, visual encoding and model building — in which visualization plays a pivotal role in enabling human insights. The first topic focuses on the visualization of large value ranges, especially in time-dependent data. Data with large value ranges are data sets whose values span several orders of magnitude. Standard visualizations, such as linear or logarithmic scales, often fail to support readability or accurate comparison across orders of magnitude. To address this, this dissertation proposes novel visualization techniques that explicitly encode both mantissa and exponent components through refined visual mappings, including a nested color scheme, a scale that bridges the strengths of linear and logarithmic axes, and multiple visual designs for single and multiple time-series data. Empirical user studies across a range of tasks — such as identification, discrimination, and estimation — demonstrate that these techniques significantly improve task accuracy, response time, and confidence in interpretation across domain-agnostic data sets. The contributions extend beyond nominal data to support complex time-series structures, for which large value ranges are common across scientific and public domains, introducing scalable designs that enhance perception of magnitude variations. All developed techniques are domain-agnostic and are practical alternatives for visualization designers facing large value ranges in time-series data. The second topic explores visual model validation, a process by which users assess the fit and plausibility of statistical or machine learning models through visual inspection. While visual estimation (i.e., visual model building) has received considerable attention, fewer studies address the perception and judgment of already computed model results. This dissertation investigates the cognitive and perceptual processes involved in validating visualized linear regression models. Through experimental human-subject studies, the accuracy of visual validation and estimation is compared, and key factors are identified that influence users’ ability to reliably validate model results. These factors include perceptual biases, user strategies, as well as data and design features. The findings contribute to a better understanding of visual validation processes and provide useful insights for machine learning applications and the design of visual analytics systems. Together, this dissertation advances the theoretical and practical foundations of visual data analysis by introducing novel techniques for encoding large value ranges and empirically evaluating human assessment of model outputs. These contributions enhance the communication of complex data in high-impact domains, such as pandemic monitoring or socioeconomic analysis. Furthermore, a deeper understanding of visual model validation enables practitioners to better interpret the uncertainty of model predictions, e.g., in medical diagnostics. The findings presented provide a foundation for future research in visualization design, human-centered AI, and explainable analytics.

Item Type:	Thesis (PhD thesis)
Creators:	Creators Email ORCID ORCID Put Code Braun, Daniel daniel.braun@uni-koeln.de orcid.org/0000-0002-8824-7184 UNSPECIFIED
URN:	urn:nbn:de:hbz:38-789742
Date:	2025
Language:	English
Faculty:	Faculty of Mathematics and Natural Sciences
Divisions:	Faculty of Mathematics and Natural Sciences > Department of Mathematics and Computer Science > Institute of Computer Science
Subjects:	Data processing Computer science
Uncontrolled Keywords:	Keywords Language Large Value Ranges UNSPECIFIED Visualization Design UNSPECIFIED Time-Dependent Data UNSPECIFIED Visual Model Validation UNSPECIFIED Regression Models UNSPECIFIED Perception UNSPECIFIED
Date of oral exam:	6 October 2025
Referee:	Name Academic Title von Landesberger, Tatiana Prof. Dr. Waldner, Manuela Prof. Dr. Andrienko, Natalia Prof. Dr.
Refereed:	Yes
URI:	http://kups.ub.uni-koeln.de/id/eprint/78974