Blockchain-Enabled Explainable Artificial Intelligence Framework for Trusted and Auditable Diabetes Risk Prediction

Resumo

Diabetes mellitus remains a major global health challenge, with its long-term complications imposing substantial economic and social burdens on healthcare systems worldwide. Although machine learning models have demonstrated strong predictive performance in identifying diabetes risk, their black-box nature severely limits clinical interpretability, trust, and accountability, thereby hindering widespread adoption in medical practice. This study introduces the Explainable Blockchain (XBC) framework, which integrates Random Forest-based diabetes risk prediction with local SHAP explanations, followed by permanent, immutable storage of concise explanation summaries on an Ethereum smart contract. Patient identifiers are pseudonymized via cryptographic hashing to preserve privacy while enabling independent auditability by clinicians, patients, and regulators. Experiments conducted on the Pima Indians Diabetes dataset yielded an accuracy of 74.68%, an AUC of 0.8118, and perfect ML-to-chain consistency (100%) across 154 test instances, with average transaction latency of 54.6 ms and gas consumption of approximately 50,595 units on a private Ganache testnet. By bridging explainable artificial intelligence and blockchain at the individual decision level, XBC establishes a verifiable, tamper-proof audit trail for AI-driven clinical decisions, significantly advancing transparency, accountability, and regulatory compliance in healthcare artificial intelligence applications.