A machine learning framework for privacy-preserving financial risk assessment using explainable artificial intelligence and blockchain-based trust management

Introduction Financial risk assessment requires models that jointly satisfy privacy, explainability, and decentralized trust, yet existing machine-learning approaches usually address these requirements in isolation. Methods This paper proposes the Privacy-Preserving Explainable Trust Framework (PP-ETF), integrating federated learning with differential privacy, post hoc SHAP and LIME explanations, and a permissioned distributed ledger for trust management. PP-ETF is evaluated on the German Credit and HELOC datasets against statistical models, centralized ensembles and deep networks, and privacy-preserving baselines. Results PP-ETF outperforms privacy-preserving baselines in AUC-ROC while remaining within 1.4%–1.6% of centralized deep-learning models. Reputation-based adversarial detection excludes a malicious participant within two communication rounds with a 3.2% false-positive rate. Explanation stability yields average cosine similarity of at least 0.87, and the ledger processes up to 480 transactions per second with 2.4 s finality latency. Discussion The results support the feasibility of jointly implementing privacy, interpretability, and decentralized trust under the evaluated conditions; they do not imply that the underlying trade-offs are eliminated. PP-ETF provides an end-to-end blueprint for effective, scalable, private, interpretable, and trustworthy risk-management systems.