Open Access

ARTICLE

Secure and Differentially Private Edge-Cloud Federated Learning Framework for Privacy-Preserving Maritime AIS Intelligence

Abuzar Khan¹, Abid Iqbal^2,*, Ghassan Husnain^1,*, Fahad Masood¹, Mohammed Al-Naeem³, Sajid Iqbal⁴

1 Department of Computer Science, CECOS University of IT and Emerging Sciences, Peshawar, Pakistan
2 Department of Computer Engineering, College of Computer Sciences and Information Technology, King Faisal University, Al-Ahsa, Saudi Arabia
3 Department of Computer Networks Communications, CCSIT, King Faisal University, Al Ahsa, Saudi Arabia
4 Department of Information Systems, College of Computer Science and Information Technology, King Faisal University, Al-Ahsa, Saudi Arabia

* Corresponding Authors: Abid Iqbal. Email: ; Ghassan Husnain. Email:

(This article belongs to the Special Issue: Cloud Computing Security and Privacy: Advanced Technologies and Practical Applications)

Computers, Materials & Continua 2026, 87(3), 21 https://doi.org/10.32604/cmc.2026.077222

Received 04 December 2025; Accepted 19 January 2026; Issue published 09 April 2026

Abstract

Cloud computing now supports large-scale maritime analytics, yet offloading rich Automatic Identification System (AIS) data to the cloud exposes sensitive operational patterns and complicates compliance with cross-border privacy regulations. This work addresses the gap between growing demand for AI-driven vessel intelligence and the limited availability of practical, privacy-preserving cloud solutions. We introduce a privacy-by-design edge-cloud framework in which ports and vessels serve as federated clients, training vessel-type classifiers on local AIS trajectories while transmitting only clipped, Gaussian-perturbed updates to a zero-trust cloud coordinator employing secure and robust aggregation. Using a public AIS corpus with realistic non-IID client partitions, our evaluation shows that non-private FedAvg attains validation AUC ≈ 0.90 and test AUC ≈ 0.78, closely matching a centralized baseline. Moderate differential privacy noise (δ≤ 0.5) preserves most of this utility across KRUM and trimmed-mean aggregation. Communication analysis indicates that secure aggregation introduces negligible overhead compared with standard FedAvg, while homomorphic encryption increases payload size by roughly an order of magnitude. Membership-inference experiments further demonstrate strong privacy protection, yielding ROC AUC ≈ 0.51 with no correctly inferred training members. Overall, the findings show that effective, regulation-conscious maritime analytics can be achieved without centralizing raw AIS data, offering a practical pathway for deploying resilient, privacy-enhanced AI services in distributed maritime environments.

---

Keywords

---