GS-BART: Bayesian Additive Regression Trees with Graph-Split Decision Rules

A 1.8 MB PDF document authored by Shuren He, last updated on April 28, 2026, introduces the GS-BART method. The paper describes a new Bayesian graph-split additive decision tree method designed to enhance performance for dependent data with graph structures. It presents a scalable informed MCMC algorithm and illustrates the method's performance in regression and classification tasks for spatial and network data.

Use Cases

• Benchmarking new graph-aware ensemble tree models against conventional methods like XGBoost and Random Forest based on the described methodology.
• Analyzing spatial dependencies in regression tasks using the graph-split decision rules that respect topology.
• Classifying network data where relationships between nodes are encoded as graph features.
• Implementing the described scalable informed MCMC algorithm for sampling decision trees in a Bayesian framework.

Strengths

• The method is specifically designed for dependent data with graph structures, a common challenge in spatial and network analysis.
• The paper includes performance comparisons against conventional ensemble tree models and Gaussian process regression models.
• The dataset is openly available under a CC-BY-4.0 license, facilitating reuse and modification.

Limitations

• The primary file is a 1.8 MB PDF, which is a paper describing a method; the availability of underlying implementation code or example data is unclear.
• Row count and column-level documentation are absent, making direct assessment of any accompanying data impossible.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

figshare

Collection Method

The dataset is a research paper proposing a methodological framework.

Freshness

Last updated 2026-04-28 15:07:01; freshness should be verified.