Author

Josiah Taylor

Date of Award

2025

Document Type

Thesis

Degree Name

Master of Science (MS)

Department

Computer Science

Committee Chair

Vineetha Menon

Committee Member

Kristin Weger

Committee Member

Jacob Hauenstein

Research Advisor

Vineetha Menon

Subject(s)

Artificial intelligence, Trust, Human-computer interaction, Explainable AI (XAI)

Abstract

Explainable Artificial Intelligence aims to open up the “black box” of artificial intelligence, thereby improving user trust, decreasing cognitive load, and improving task performance. As the number of machine learning models has grown, so has the number of explanation methods, making it necessary to select the most suitable explanation method for a given context. However, current selection methods often fail to account for all the factors that contribute significantly to an explanation method’s suitability. This thesis proposes a novel methodology for selecting and evaluating explanation methods. Subjective workload, objective performance, and feasibility are considered. This three-pronged “pitchfork” methodology is put to use by evaluating the performance of LIME and Kernel SHAP in a simulated environment where users are tasked with rescuing people. Quantitative data relating to subjective workload, objective performance, and feasibility is captured and analyzed in order to determine the most suitable explanation for the context.

Available for download on Monday, June 15, 2026

Share

COinS