Federated learning intrusion detection system for internet of things / Yacine Chettouh

Public ISBD Titre : Federated learning intrusion detection system for internet of things Type de document : document multimédia Auteurs : Yacine Chettouh, Auteur ; Mohamed Houssam Eddine Guenou, Auteur ; Yousra Cheriguene, Directeur de thèse Editeur : Laghouat : Université Amar Telidji - Département d'informatique Année de publication : 2026 Importance : 55 p. Accompagnement : 1 disque optique numérique (CD-ROM) Note générale : Option : Networks, systems and distributed applications Langues : Anglais (eng) Résumé : Nowadays, the Internet of Things (IoT) is widely used across critical infrastructures, connecting billions of embedded edge devices. In this context, securing these decentralized networks faces significant limitations due to the severe resource constraints and privacy requirements of IoT nodes, making traditional centralized Intrusion Detection Systems (IDS) impractical. While Federated Learning (Federated Learning (FL)) offers a privacy-preserving alternative, standard protocols like FedAvg face problems such as severe network congestion and high communication overhead caused by the continuous transmission of massive, uncompressed model updates over low-bandwidth links. To address these problems, therefore, this dissertation proposes a new communication-efficient Federated Learning Intrusion Detection System (FL-IDS). Leveraging a FedProx optimization core, the framework integrates an on-device Adaptive Gradient Quality Gate and top-k parameter sparsification to filter redundant updates and drastically reduce payload sizes. Furthermore, an emergency staleness override guarantees stable participation across the heterogeneous network. The performance of the proposed architecture is analyzed thoroughly and extensively through simulation using the CICIoT2023 dataset. The results demonstrate that our framework achieves a massive 80% reduction in cumulative communication bandwidth compared to standard FedAvg, while successfully maintaining a high global detection accuracy of 91.5% and a macro F1-score of 0.786. Hence, this work provides a significant improvement in IoT security protocols, presenting a reliable, resource-aware solution for modern edge environments while establishing a basis for future research in the field. note de thèses : Mémoire de master en informatique Federated learning intrusion detection system for internet of things [document multimédia] / Yacine Chettouh, Auteur ; Mohamed Houssam Eddine Guenou, Auteur ; Yousra Cheriguene, Directeur de thèse . - Laghouat : Université Amar Telidji - Département d'informatique, 2026 . - 55 p. + 1 disque optique numérique (CD-ROM). Option : Networks, systems and distributed applications Langues : Anglais (eng) Résumé : Nowadays, the Internet of Things (IoT) is widely used across critical infrastructures, connecting billions of embedded edge devices. In this context, securing these decentralized networks faces significant limitations due to the severe resource constraints and privacy requirements of IoT nodes, making traditional centralized Intrusion Detection Systems (IDS) impractical. While Federated Learning (Federated Learning (FL)) offers a privacy-preserving alternative, standard protocols like FedAvg face problems such as severe network congestion and high communication overhead caused by the continuous transmission of massive, uncompressed model updates over low-bandwidth links. To address these problems, therefore, this dissertation proposes a new communication-efficient Federated Learning Intrusion Detection System (FL-IDS). Leveraging a FedProx optimization core, the framework integrates an on-device Adaptive Gradient Quality Gate and top-k parameter sparsification to filter redundant updates and drastically reduce payload sizes. Furthermore, an emergency staleness override guarantees stable participation across the heterogeneous network. The performance of the proposed architecture is analyzed thoroughly and extensively through simulation using the CICIoT2023 dataset. The results demonstrate that our framework achieves a massive 80% reduction in cumulative communication bandwidth compared to standard FedAvg, while successfully maintaining a high global detection accuracy of 91.5% and a macro F1-score of 0.786. Hence, this work provides a significant improvement in IoT security protocols, presenting a reliable, resource-aware solution for modern edge environments while establishing a basis for future research in the field. note de thèses : Mémoire de master en informatique

---

1 (1 - 1 / 1) Par page : 25 50 100 200