Comparing Visual Encodings for the Task of Anomaly Detection

Meirav Taieb-Maimon, Eden Ya’akobi, Nevo Itzhak, Yossi Zaltsman

Research output: Contribution to journalArticlepeer-review

Abstract

Empirical evaluations of visual encodings for analytical tasks inform the design of automatic presentation systems. This study compares anomaly detection effectiveness, efficiency, and user satisfaction using tabular visualization and graphical representation of position, size, and color saturation visualizations and the visualizations’ relative ranking. In our user study analysts used the visualizations to detect anomalies in bivariate quantitative, ordinal, and nominal real data, before and after training. Consistent with Mackinlay ranking of the visual encodings’ effectiveness, the results showed that for all data types, position visualization outperformed size and color saturation visualizations on all measures, before and after training. The use of position visualization after training was as effective as using the table visualization for all data types but significantly easier to use and preferable. Furthermore, the average anomaly detection time was at least three times shorter when using position visualization compared to table visualization for ordinal and quantitative data.

Original languageEnglish
JournalInternational Journal of Human-Computer Interaction
DOIs
StateAccepted/In press - 1 Jan 2022

ASJC Scopus subject areas

  • Human Factors and Ergonomics
  • Human-Computer Interaction
  • Computer Science Applications

Fingerprint

Dive into the research topics of 'Comparing Visual Encodings for the Task of Anomaly Detection'. Together they form a unique fingerprint.

Cite this