Graph-Based Fraud Detection with Dual-Path Graph Filtering

2026-04-14 · Source: Machine Learning · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics · Depth: Expert, quick

Summary

A new Graph-Based Fraud Detection Model with Dual-Path Graph Filtering (DPF-GFD) has been proposed to enhance fraud detection on graph data, a task where traditional Graph Neural Networks (GNNs) often underperform due to relation camouflage, high heterophily, and class imbalance. DPF-GFD employs a novel frequency-complementary dual-path filtering paradigm. It first uses a beta wavelet-based operator on the original graph to capture structural patterns, then constructs a similarity graph from distance-based node representations, applying an improved low-pass filter. The embeddings from both graphs are fused via supervised representation learning, and the resulting node features are fed into an ensemble tree model to assess fraud risk. This approach explicitly decouples structural anomaly modeling and feature similarity modeling, leading to more discriminative and stable node representations, as demonstrated by experiments on four real-world financial fraud detection datasets.

Key takeaway

For research scientists developing fraud detection systems, DPF-GFD offers a robust approach to mitigate challenges posed by heterophilous and imbalanced graph data. You should consider integrating its dual-path filtering paradigm, which explicitly separates structural anomaly and feature similarity modeling, to achieve more stable and discriminative node representations in your models, potentially improving detection accuracy on real-world financial datasets.

Key insights

DPF-GFD uses dual-path graph filtering to overcome GNN limitations in fraud detection on heterophilous and imbalanced graphs.

Principles

• Decouple structural and feature similarity modeling.
• Combine wavelet and low-pass filtering for graph data.

Method

DPF-GFD applies a beta wavelet operator to the original graph, constructs a similarity graph with a low-pass filter, fuses embeddings, and uses an ensemble tree for fraud risk assessment.

In practice

• Apply dual-path filtering for heterophilous graphs.
• Use ensemble trees for final fraud risk assessment.

Topics

• Graph-Based Fraud Detection
• Dual-Path Graph Filtering
• Graph Neural Networks
• Heterophily
• Ensemble Tree Models

Best for: Research Scientist, AI Scientist, Machine Learning Engineer, Data Scientist

Related on AIssential

• TMR-GGNN: Credit Card Fraud Detection based on Time-Aware Multi-Relational Guided Graph Neural Network — TMR-GGNN leverages time-aware multi-relational GNNs and contrastive learning to improve credit card fraud detection, especially for rare cases.
• A Zero-shot Generalized Graph Anomaly Detection Framework via Node Reconstruction — AlignGAD enables zero-shot graph anomaly detection across heterogeneous domains by unifying features and utilizing multi-scale cluster-aware views.
• Two-Stage Learned Decomposition for Scalable Routing on Multigraphs — NEPF decomposes multigraph VRPs into node permutation and edge selection for scalable, high-quality routing.
• Learning Long Range Spatio-Temporal Representations over Continuous Time Dynamic Graphs with State Space Models — CTDG-SSM unifies state-space modeling with graph topology to capture long-range spatio-temporal dependencies in continuous-time dynamic graphs.
• GDGU: A Gradient Difference-based Graph Unlearning Method for Cyberattack Localization in Electric Vehicle Charging Networks — GDGU efficiently unlearns specific data from GNNs for EVCS cyberattack localization, matching utility with faster execution.
• GJDNet: Robust Graph Neural Networks via Joint Disentangled Learning Against Adversarial Attacks — GJDNet enhances GNN robustness by jointly disentangling node representations and decision spaces against adversarial attacks.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Machine Learning.