Random-Forest-Induced Graph Neural Networks for Tabular Learning

TL;DR

This paper introduces RF-GNN, a novel framework that constructs graphs from tabular data using random forest proximities, enabling effective application of GNNs to improve tabular learning performance.

Contribution

RF-GNN is the first method to create instance-level graphs from tabular data via random forest proximities, facilitating GNN use without restrictive assumptions.

Findings

RF-GNN outperforms classical baselines in weighted F1-score across 36 datasets.

Proximity design choices significantly influence GNN performance.

Graph construction settings impact the effectiveness of RF-GNN.

Abstract

Graphs are essential for modeling complex relationships and capturing structured interactions in data. Graph Neural Networks (GNNs) are particularly effective when such relational structure is explicitly available, but many real-world datasets, most notably tabular data, lack an inherent graph representation. To address this limitation, we propose RF-GNN, a framework that constructs instance-level graphs from tabular data using proximity measures induced by random forests. These proximities capture nonlinear feature interactions and data-adaptive similarity without imposing restrictive assumptions on feature geometry. The resulting graphs enable the direct application of GNNs to tabular learning problems. Extensive experiments on 36 benchmark datasets demonstrate that RF-GNN consistently outperforms strong classical baselines and recent graph-construction methods in terms of weighted F1-score. Additional ablation studies highlight the impact of proximity design choices and graph construction settings.