Graph Classification via Simple Graph Based Features
Farida Abdelmoneum, John Shi, José M. F. Moura · 2023
Graph classification is typically done with geometric deep learning. Given a set of graphs from different classes, e.g., representing mutagenic and non-mutagenic molecules in the MUTAG dataset, and data defined on the graphs, geometric deep learning architectures such as graph convolutional neural networks and graph attention networks can be used to classify graphs into a set of classes. In this paper, we propose a statistical approach to classify graphs. Unlike deep learning architectures, we design graph features that reflect the graph structure and then build a classifier based on the features. The paper provides an ablation study of our algorithm on various traditional machine learning algorithms and graph classification datasets.