Privacy-preserving federated analysis of agricultural loan and insurance risk

A robust agricultural loan and insurance risk analysis is conducted on sensitive data aggregated from multiple sources, such as banks, insurers, farmers, and agritech firms. This research integrates privacy-preserving federated analysis with quantum-resistant lattice-based fully homomorphic encryption (FHE) to ensure compliance with data regulations and mitigate exposure risks. The methodology enables secure multi-party computation directly on encrypted data with ML algorithms (XGBoost, neural networks). It was tested by NASA satellite images processed by IBM’s Prithvi Large Language Model (LLM) with encrypted ML inferences on the XGBoost decision tree and neural network. Comparative evaluation against unencrypted data reveals minimal accuracy loss, demonstrating practical viability without compromising predictive performance.