Research map / Label-efficient and robust learning / Recognition and long-tail
Evaluating the Effectiveness of Explainable Artificial Intelligence Approaches (Student Abstract)
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Focus-group study with seven graduate students ranking four XAI explanation methods (XGBoost feature importance, SHAP, LIME, permutation importance) for an obesity prediction model on interpretability and plausibility.
Problem
Explanation effectiveness of XAI approaches is inadequately evaluated in healthcare, with no standardized evaluation framework; it is unclear how subjective interpretability relates to perceived plausibility and what drives users' acceptance of explanations.
Approach
Uses a qualitative focus group to rank four XAI explanation techniques applied to an XGBoost obesity classifier on subjective interpretability and plausibility, and categorizes the rationales (intuitiveness, scope, mechanism, contrastive) to identify what drives acceptance.
Key results
- XGBoost feature importance was rated the most plausible explanation, followed by SHAP and LIME; rankings for interpretability and plausibility were mostly consistent except for SHAP.
- Intuitiveness of the explanation was the most common justification for interpretability ranking, followed by scope, mechanism, and contrastive explanation.
- Authors conclude that easily interpreted explanations are perceived as more plausible, though generalizability is limited by the small focus group of seven students.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use XGBoost feature importance rather than the other evaluated explanations | XAI explanations for XGBoost obesity prediction · KNHANES obesity prediction focus group | perceived plausibility | Helped |
| Use SHAP rather than LIME | XAI explanations for XGBoost obesity prediction · KNHANES obesity prediction focus group | perceived plausibility | Helped |
| Make explanations more intuitive | XAI explanations · KNHANES obesity prediction focus group | subjective interpretability | Helped |
| Provide a contrastive “why not” explanation | XAI explanations · KNHANES obesity prediction focus group | understanding | Helped |
| Include too many predictors in the explanation | XAI explanations · KNHANES obesity prediction focus group | user distraction | Hurt |
| Improve subjective interpretability | XAI explanations · KNHANES obesity prediction focus group | perceived plausibility | Helped |
| Vary the XAI explanation technique | XAI explanations for XGBoost obesity prediction · KNHANES obesity prediction focus group | ranking consistency between interpretability and plausibility | Mixed |
Surprising findings
- Contrastive ('why not') explanations were mentioned by only two focus group participants despite being viewed as helpful.
Limitations
- Small focus group of seven graduate students limits generalizability.
- Evaluation relies on subjective rankings and qualitative rationales rather than objective measures of explanation effectiveness.
Experimental setup
Datasets: Korean National Health and Nutrition Examination Survey (KNHANES)
Metrics: average ranking score for interpretability average ranking score for plausibility frequency of rationale categories
Builds on: SHAP LIME XGBoost
Closest papers by meaning
- Position: Explain to Question not to Justify (ICML 2024 · similarity 0.90)
- Navigating the Maze of Explainable AI: A Systematic Approach to Evaluating Methods and Metrics (NeurIPS 2024 · similarity 0.87)
- Position: Explainable AI Cannot Advance Without Better User Studies (ICML 2025 · similarity 0.85)
- iLLuMinaTE: An LLM-XAI Framework Leveraging Social Science Explanation Theories Towards Actionable Student Performance Feedback (AAAI 2025 · similarity 0.83)
- Explainability matters: The effect of liability rules on the healthcare sector (NeurIPS 2026 · similarity 0.82)
- Semi-factual Explanations in AI (AAAI 2024 · similarity 0.82)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.