Development and Evaluation of Two Approaches of Visual Sensitivity Analysis to Support Epidemiological Modeling

Area of Science:

• Computational science
• Data visualization
• Scientific modeling

Background:

• Computational modeling is vital across disciplines, notably in combating the COVID-19 pandemic.
• Sensitivity analysis is crucial for monitoring model behavior but traditional methods lack insight into parameter interactions.
• Current methods require extensive model runs for actionable parameter optimization insights.

Purpose of the Study:

• To develop and compare two visual analytics approaches for enhanced sensitivity analysis in computational modeling.
• To address limitations in understanding parameter interactions and optimizing model parameters.
• To provide actionable insights for model development and deployment.

Main Methods:

• Developed two visual analytics approaches: algorithm-centric and visualization-assisted, and visualization-centric and algorithm-assisted.
• Evaluated approaches through structured task analysis for visual sensitivity analysis.
• Incorporated domain expert feedback for validation.

Main Results:

• The developed visual analytics approaches offer improved understanding of parameter interactions compared to traditional methods.
• Both algorithm-centric and visualization-centric approaches provide distinct advantages for sensitivity analysis.
• The methods facilitate more efficient parameter optimization through better data interpretation.

Conclusions:

• Visual analytics significantly enhances sensitivity analysis in computational modeling.
• The presented approaches are versatile and applicable to various modeling processes with time-series outputs.
• These methods can be extended to models with different output types, broadening their impact.