Privacy-Preserving Federated Mining of Utility-Based Patterns

Abstract

Collaborative high-utility pattern mining (HUPM) aims to identify high-utility patterns (HUPs) across multiple data owners, and it is particularly suitable for distributed Internet of Things (IoT) scenarios such as smart retail and edge-assisted service analytics. However, collaborative mining raises privacy concerns, as sensitive local records and business statistics may be exposed during the mining process. To address this problem, this article presents FLUM, a privacy-preserving federated framework for HUPM based on participant-side differential privacy (DP). FLUM follows an interactive query–response process in which a fixed group of participants locally computes the utility, transaction-weighted utilization (TWU), and remaining-utility upper bound (REUB) of candidate patterns, and uploads only perturbed statistics. In addition, FLUM incorporates adaptive sensitivity computation, bitmap-based local acceleration, and two candidate-generation strategies to reduce the search space of HUPM. Both theoretical analysis and experimental results show that FLUM can protect privacy while still achieving competitive utility and efficiency. The source code is available on GitHub: https://github.com/CSYaoChen/FLUM