Extracting Interpretable Models from Tree Ensembles: Code and Experimental Data

Brian Liu's research on figshare, last updated April 28, 2026, proposes an estimator for extracting compact decision rules from tree ensembles. The 12.5 MB repository includes code and data supporting the development of exact and approximate algorithms for rule extraction. The work establishes non-asymptotic prediction error bounds and demonstrates performance against existing algorithms through experiments.

Use Cases

• Benchmarking rule extraction algorithms based on the described estimator and optimization methods.
• Studying the trade-off between model complexity and predictive accuracy for interpretable models.
• Reproducing experiments that compare extracted rule sets to original tree ensemble performance.

Strengths

• Includes source code and experimental data in a 12.5 MB repository, facilitating reproducibility.
• Released under the permissive CC-BY-4.0 license, allowing for broad reuse and modification.

Limitations

• Row count and specific column definitions are unknown, limiting suitability assessment.
• Column-level documentation is absent; field semantics must be inferred after download.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

figshare

Collection Method

Likely generated as part of academic research on machine learning interpretability.

Freshness

Last updated 2026-04-28 15:10:24; freshness should be verified.