Hybrid LSTM-Graph Neural Framework for Robust Financial Fraud Detection and Adversarial Resilience

Short summary

FraudShield AI combines LSTM networks with graph topological features to detect sophisticated money laundering patterns like smurfing and layering, addressing extreme class imbalance (0.13% fraud rate) and adversarial evasion. The framework shifts from isolated transaction analysis to network-level forensics using PageRank centrality, in-degree dynamics, and a custom flow ratio. On the PaySim dataset, it outperforms Logistic Regression and XGBoost baselines in precision, recall, and F1-score, especially for micro-transaction fraud.

• •Hybrid LSTM + graph feature framework for detecting money laundering patterns like smurfing and layering
• •Uses Focal Loss and dynamic thresholding to handle 0.13% fraud rate and adversarial evasion
• •Outperforms XGBoost and Logistic Regression baselines on PaySim dataset, particularly on micro-transaction fraud

Generated with AI, which can make mistakes.