Frequent Subgraph-based Persistent Homology for Graph Classification

This paper introduces a novel graph filtration method, Frequent Subgraph Filtration (FSF), to improve graph classification by leveraging persistent homology. It addresses the limitations of existing methods that rely on simpler filtrations by incorporating richer features from frequent subgraphs. The paper proposes two classification approaches: an FPH-based machine learning model and a hybrid framework integrating FPH with graph neural networks. The results demonstrate competitive or superior accuracy compared to existing methods, highlighting the potential of FSF for topology-aware feature extraction in graph analysis.

• •Proposes Frequent Subgraph Filtration (FSF) for graph classification.
• •Introduces FPH-ML and FPH-GNNs for graph classification.
• •FSF improves performance compared to existing methods.
• •Hybrid framework with GNNs shows significant gains.