Statistical Comparison of Explanation Faithfulness Metrics

369 paired statistical comparisons of explanation faithfulness metrics, specifically per-image AUCs, for two AI models. Md Mehedi Hasan Santo published this dataset on figshare in April 2026. It contains mean differences and p-values from paired t-tests and Wilcoxon signed-rank tests.

Use Cases

• Compare the faithfulness of MaxGRNet versus a GaussianBlur baseline using reported mean differences and p-values.
• Validate statistical significance of model performance differences using paired t-test and Wilcoxon signed-rank test results.
• Analyze per-image AUC scores for Insertion and (1-Deletion) metrics across 369 common samples.

Strengths

• Contains 369 paired observations, providing a substantial sample for statistical analysis.
• Includes results from two established statistical tests (paired t-test and Wilcoxon signed-rank test).
• Published under the open CC-BY-4.0 license.

Limitations

• The dataset is very small, only 5.5 KB, indicating limited scope and likely a single table of summary statistics.
• No raw per-image AUC data or column names are provided, only aggregated test results.
• The temporal coverage and geographic scope of the underlying image data are unknown.

Provenance

Source

Md Mehedi Hasan Santo via figshare.

Collection Method

Statistical comparison derived from per-image AUCs calculated on a common subset of 369 images.

Freshness

Last updated in April 2026.