Detecting Anomalies in Blockchain Transactions using Machine Learning Classifiers and Explainability Analysis

Explainability is critical for trust in AI-driven financial security: The paper addresses a significant gap in blockchain anomaly detection by integrating explainable AI (XAI) techniques, specifically SHAP analysis, to show why the system flags transactions as suspicious rather than just providing a yes/no prediction—essential for users and regulators to trust automated financial systems.

Imbalanced data requires specialized solutions: Since fraudulent transactions are rare compared to legitimate ones, the researchers developed a new algorithm called XGBCLUS to better balance training data. This addresses a common real-world challenge in responsible AI: ensuring detection systems don't become biased toward the majority class at the expense of catching actual anomalies.

Ensemble methods outperform single models: The study found that combining multiple machine learning classifiers (stacking and voting approaches) produced better results than individual models, suggesting that responsible AI systems for high-stakes applications like financial security benefit from diverse analytical approaches rather than relying on a single decision-maker.

Interpretable rules make AI decisions actionable: Beyond statistical explanations, the paper presents human-readable rules for determining whether transactions are anomalous, making the system's logic transparent and auditable—a key requirement for responsible deployment in regulated financial environments.