Explainable Artificial Intelligence Framework for Blockchain Transaction Risk Prediction Using Behavioral Indicators and Transaction Patterns

Blockchain technology has become a fundamental infrastructure for decentralized financial systems and digital asset transactions. However, the increasing volume and complexity of blockchain transactions have introduced significant security challenges, including the identification of potentially risky or suspicious activities. This study proposes an Explainable Artificial Intelligence (XAI) framework for blockchain transaction risk prediction using behavioral indicators and transaction patterns. The proposed framework integrates Extreme Gradient Boosting (XGBoost) for classification, Synthetic Minority Oversampling Technique (SMOTE) for addressing class imbalance, and SHapley Additive exPlanations (SHAP) for model interpretability. The experiments were conducted on a blockchain transaction risk dataset containing 9,000 transaction records with behavioral features such as transaction amount, transaction fee, wallet age, transaction velocity, mixing service usage, cross-border activity, and darknet association indicators. The results show that SMOTE effectively balanced the dataset and improved the detection of minority-class transactions. The proposed model achieved an accuracy of 68.17%, precision of 18.45%, recall of 30.82%, F1-score of 23.09%, and ROC-AUC of 52.67%. Although the predictive performance remains limited, the explainability analysis provides valuable insights into the factors influencing transaction risk. Feature importance analysis identified mixing_service_used and cross_border_flag as the most influential variables, while SHAP analysis highlighted unusual_pattern_score, avg_wallet_transaction_value, and total_wallet_transactions as the most impactful predictors. The findings demonstrate the usefulness of explainable artificial intelligence for understanding blockchain transaction risk and identifying key behavioral indicators associated with suspicious activities. The proposed framework contributes to blockchain security research by enhancing transparency in machine learning-based risk assessment and providing insights for future development of intelligent blockchain monitoring systems.