Safety-Relevant Error Rates for XGBoost and Fine-Tuned Models on 14,832 Test Samples

14,832 test samples form the basis for comparing safety-relevant error rates between a structured-data XGBoost comparator and a fine-tuned model. The dataset, authored by Eui Jun Lee, was last updated on June 4, 2026. It is a small, 5.5 KB Excel file shared under a CC-BY-4.0 license on figshare.

Use Cases

• Compare safety error rates between XGBoost and fine-tuned models based on the described test subset.
• Benchmark model safety performance on a structured-data task.
• Analyze the impact of fine-tuning on safety-relevant metrics.
• Validate error rate claims for models mentioned in the description.

Strengths

• Dataset is small (5.5 KB), enabling quick download and inspection.
• The test subset size is explicitly defined as 14,832 samples.
• It has a clear, permissive license (CC-BY-4.0).

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count for the full dataset is unknown, which may limit suitability assessment.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

figshare

Freshness

Last updated 2026-06-04 17:25:22; freshness should be verified.