Fea2Fea: Exploring Structural Feature Correlations via Graph Neural Networks

TL;DR

This paper introduces a graph neural network-based method to analyze correlations between structural features in geometric graphs, demonstrating improved classification accuracy through feature filtering and combination strategies.

Contribution

It pioneers the use of GNNs for structural feature correlation analysis and shows how feature filtering enhances graph classification tasks.

Findings

High correlation exists between some structural features.

Filtered feature combinations improve classification accuracy.

Simple concatenation methods outperform complex ones.

Abstract

Structural features are important features in a geometrical graph. Although there are some correlation analysis of features based on covariance, there is no relevant research on structural feature correlation analysis with graph neural networks. In this paper, we introuduce graph feature to feature (Fea2Fea) prediction pipelines in a low dimensional space to explore some preliminary results on structural feature correlation, which is based on graph neural network. The results show that there exists high correlation between some of the structural features. An irredundant feature combination with initial node features, which is filtered by graph neural network has improved its classification accuracy in some graph-based tasks. We compare differences between concatenation methods on connecting embeddings between features and show that the simplest is the best. We generalize on the synthetic geometric graphs and certify the results on prediction difficulty between structural features.