XBlock-ExplainNet: An Interpretable Deep Learning Model for Fraud Detection in Blockchain-Based Fintech Systems

s the fintech industry has evolved to use blockchain technology rapidly, fraudulent transactions and fraud using blockchains have become a significant issue to consider as a result of the decentralization category of blockchain networks which are pseudonymous. The typical machine learning models cannot deal efficiently with graphstructured data in the blockchain and tend to operate as black boxes, which cannot be applied to the regulated financial setting. Drawing on graph-based learning, temporal modeling, and explainable AI, we in this work introduce the XBlockExplain-Net, which is an interpretable deep learning framework to detect the fraud effectively. The architecture suggested has a two stream feature extractor which processes graphs and transaction sequences independently. They are integrated and provided to a Graph-Aware Temporal Transformer Encoder (GTTE) that initiates blockchain type with self-attention operations. In order to solve the problem of transparency, an Explainable Fusion Layer has been presented which combines SHAP feature attribution and internal attention scores to provide transparency to the decision-making done. Outputting of final predictions with confidence scores is done by using a residual classifier. The Elliptic Bitcoin Dataset was used to provide experimental validation of the model, which exhibited an accuracy of 95.6%, F1-score of 94.0, and a ROC-AUC score of 0.96, which were far better than the traditional methods, including GCN and GAT. The explainability score of the framework was also quite high (0.93), proving the lack of opacity and the model applicability to real-world. The findings are sufficient to corroborate that XBlock-ExplainNet presents a novel benchmark to fraud detection within blockchain fintech systems through a balance between interpretability and accuracy.