DDoS_FL: Federated learning architecture approach against DDoS attack

Authors : Büşra Büyüktanir, Zeki Çıplak, Abdullah Emir Çil, Özlem Yakar, Mahamoud Brahim Adoum, Kazım Yıldız

Pages : 1004-1018

Doi:10.5505/pajes.2025.40456

View : 80 | Download : 283

Publication Date : 2025-11-13

Article Type : Research Paper

Abstract :The frequency and complexity of DDoS attacks have significantly increased with the growth of the internet, posing severe threats to network security. Traditional machine learning and deep learning-based detection systems often face limitations due to their reliance on centralized data collection, leading to privacy concerns, high computational costs, and challenges in adapting to heterogeneous data distributions. This study proposes DDoS_FL, a federated learning-based model designed to detect DDoS attacks without requiring data sharing between devices. The model has demonstrated effectiveness under both Independent and Identically Distributed (IDD) and Non-Independent and Identically Distributed (Non-IDD) data distributions while preserving data privacy and maintaining high detection accuracy. The proposed model is trained and evaluated using the CIC-DDoS2019 dataset, which includes various types of DDoS attacks. Experimental results show that federated learning significantly reduces training time compared to traditional centralized approaches while achieving detection accuracy ranging from 82% to 97%. Furthermore, the scalability of the model is analyzed based on the number of participating clients, highlighting the advantages of its distributed nature. Comparative analyses confirm that the proposed approach is competitive in both privacy preservation and detection performance. This study demonstrates that federated learning provides an effective solution for detecting DDoS attacks and has significant potential in enhancing network security.
Keywords : DDoS saldırısı, Federe öğrenme, Veri gizliliği, Derin sinir ağı, Derin öğrenme, Bilgi güvenliği

ORIGINAL ARTICLE URL