TREE-G: Decision Trees Contesting Graph Neural Networks
This addresses the problem of effectively incorporating topological information into decision trees for graph data, offering improved accuracy and explainability for researchers and practitioners in graph learning, though it is an incremental advancement building on existing tree-based and graph-learning methods.
The paper tackled the challenge of applying decision trees to graph-structured data by introducing TREE-G, a novel method that modifies standard decision trees with a graph-specialized split function incorporating node features, topology, and a pointer mechanism, resulting in consistent outperformance of other tree-based models and often Graph Neural Networks on multiple benchmarks, sometimes by large margins.
When dealing with tabular data, models based on decision trees are a popular choice due to their high accuracy on these data types, their ease of application, and explainability properties. However, when it comes to graph-structured data, it is not clear how to apply them effectively, in a way that incorporates the topological information with the tabular data available on the vertices of the graph. To address this challenge, we introduce TREE-G. TREE-G modifies standard decision trees, by introducing a novel split function that is specialized for graph data. Not only does this split function incorporate the node features and the topological information, but it also uses a novel pointer mechanism that allows split nodes to use information computed in previous splits. Therefore, the split function adapts to the predictive task and the graph at hand. We analyze the theoretical properties of TREE-G and demonstrate its benefits empirically on multiple graph and vertex prediction benchmarks. In these experiments, TREE-G consistently outperforms other tree-based models and often outperforms other graph-learning algorithms such as Graph Neural Networks (GNNs) and Graph Kernels, sometimes by large margins. Moreover, TREE-Gs models and their predictions can be explained and visualized