Design of an Integrated Modern Explainable Machine Learning Framework for Blockchain Forensic Analysis

บทคัดย่อ

The rapid expansion of blockchain technology has led to a surge in fraud cases, demanding advanced forensic methods to ensure security and transparency. Traditional static, rule-based models are inadequate for the complex and dynamic nature of blockchain transactions, while existing graph-based anomaly detection methods still struggle with temporal awareness, adaptability, and cross layer integration. These models also suffer from high false positives, inefficient thresholding, and limited explainability, reducing their effectiveness in real-world investigations. To address these gaps, we propose a Modern Explainable Machine Learning Framework for Blockchain Forensic Analysis, integrating five advanced AI techniques to enhance both fraud detection and interpretability. The Temporal Graph Transformer (TGT) identifies evolving transaction patterns with 96.5% accuracy, while Reinforcement Learning-Based Adaptive Fraud Thresholding (RL-FT) reduces false positives by 45%. Contrastive Self-Supervised Blockchain Embedding (CSBE) improves fraud separation by 30%, and Hybrid Diffusion-Based Anomaly Forecasting (HDAF) predicts future fraud with 94.1% accuracy. Finally, Multi-Modal Blockchain Forensic Fusion (MBFF) combines transactional, smart contract, and network data for 99.1% detection accuracy and 50% better explainability. This integrative forensic intelligence system effectively overcomes key limitations in current blockchain fraud analysis.