Homomorphic Encryption and Secure Multi-Party Computation for Privacy-Preserving Data Mining in Banking

Abstract

Abstract
The banking sector increasingly relies on data mining and machine learning across distributed datasets to perform credit scoring, fraud detection, anti-money-laundering (AML) analytics, and personalized services. These capabilities, however, are constrained by stringent privacy requirements, regulatory obligations, and the commercial sensitivity of customer data. Cryptographic primitives principally Homomorphic Encryption (HE) and Secure Multi-Party Computation (SMPC, also MPC) provide mathematically grounded approaches to compute on private data without revealing underlying raw inputs. This manuscript synthesizes theory, system architectures, protocol choices, and applied patterns for deploying HE and SMPC in banking data-mining workflows. This paper (1) review the mathematical foundations and practical HE schemes (BFV, BGV, CKKS, TFHE, Paillier) and dominant MPC paradigms (Yao, GMW, SPDZ, garbled circuits, secret sharing); (2) evaluate performance, precision, and communication tradeoffs using current library ecosystems (Microsoft SEAL, HElib, OpenFHE) and MPC frameworks; (3) present reference architectures and hybrid HE–MPC compositions for realistic banking tasks (fraud detection, collaborative AML, privacy-preserving model training and inference, private set intersection); (4) propose evaluation metrics, threat models, and compliance considerations; and (5) identify research directions for scalability, latency, verifiability, and regulatory alignment.