FedHoeffCrypt: A Server-Side Dynamic Watermarking Scheme With Modified Hoeffding Homomorphic Encrypted Federated Learning Algorithm for IoT-IDS

• Suma S Girija,
• Rukmani Panjanathan,

Vol. 20, No. 1, January 31, 2026
10.3837/tiis.2026.01.020, Download Paper (Free):

• Federated learning
• homomorphic encryption
• Intrusion Detection System
• IoT
• Model Watermarking.

Abstract

Federated learning (FL) can offer an attractive solution to the problem of training overhead related to large datasets in the context of Internet of Things-based Intrusion Detection Systems (IoT-IDS). However, FL can be susceptible to a number of adversarial attacks because of its client-server architecture in which aggregated model updates can disclose sensitive parameters. In order to overcome these vulnerabilities, homomorphic encryption (HE) is often used to achieve secure aggregation at the server level. Despite this, the diverse parameters of the deep/machine learning models are vulnerable to unauthorized access during the training process, therefore there is need to apply model watermarking as the means of protecting intellectual property. The primary complexity lies in the implementation of the server-side watermarking in an encrypted framework since watermarking is typically incorporated in non-encrypted parameters. To address these issues, this paper proposes a secure FL-based Intrusion Detection System (IDS) that integrates HE with a new server-side dynamic watermarking scheme. Furthermore, the system includes a variant of Hoeffding Anytime Tree (HAT) algorithm that is trained to operate in an FL setup to determine the anomalies and make significant features to categorize the attacks. The dynamic watermarking mechanism ensures that the security of the client and the server is not compromised. The client does not need to exchange any private information with the server, which makes use of the efficient construction of the trigger set. The performance of the proposed system was evaluated using the CICIDS2017, CICIDS2018, and the EdgeIIoT data sets. The proposed system is superior to the existing models as it delivers greater accuracy, strong data privacy at lower computational cost.

Statistics

Cite this article