MARFD: Multi-Layered Adversarial Machine Learning Framework for Fraud Detection and Threat Modeling in Decentralized Finance

The MARFD framework integrates unsupervised anomaly detection, ensemble classification, adversarial robustness training, explainability (SHAP/LIME-compatible permutation importance), and STRIDE–MITRE threat mapping into a unified architecture for secure and interpretable DeFi fraud detection. This archive includes: Fully executable Python codebase (MARFD v1.0) Synthetic DeFi-style dataset (synthetic_transactions.csv, 20,000 samples) Preprocessing, evaluation, and threat-mapping scripts Results folder with example outputs (metrics JSON, feature importances, threat logs) Documentation (README.md and requirements.txt) All files are released under MIT License and are completely reproducible without restrictions.The dataset contains no human or sensitive information and was programmatically generated to emulate DeFi transaction behavior patterns.