Geri Dön

Explainable federated learning for IOT security systems: Balancing privacy, transparency, and performance

Explaınable federated learnıng for ıot securıty systems: balancıng prıvacy, transparency, and performance

  1. Tez No: 1005541
  2. Yazar: RODAN ALSALEH
  3. Danışmanlar: YRD. DOÇ. DR. AHMED ŞENOL
  4. Tez Türü: Yüksek Lisans
  5. Konular: Bilim ve Teknoloji, Bilgisayar Mühendisliği Bilimleri-Bilgisayar ve Kontrol, Science and Technology, Computer Engineering and Computer Science and Control
  6. Anahtar Kelimeler: Belirtilmemiş.
  7. Yıl: 2025
  8. Dil: İngilizce
  9. Üniversite: Üsküdar Üniversitesi
  10. Enstitü: Fen Bilimleri Enstitüsü
  11. Ana Bilim Dalı: Siber Güvenlik Ana Bilim Dalı
  12. Bilim Dalı: Belirtilmemiş.
  13. Sayfa Sayısı: Belirtilmemiş.

Özet

There is an accelerating trend for governments to adopt digital transformation, which is clearly reflected in the growing presence of Internet of Things (IoT) deployments worldwide. These technologies have significantly enhanced the quality and performance of various services. However, this rapid increase in the adoption of the IoT has also raised significant security and privacy concerns, necessitating the development of new technological solutions to secure IoT systems. In these solutions, Intrusion Detection Systems (IDSs) play a crucial role. AI-powered IDSs have been developed as effective real-time intrusion detection and response solutions. However, the dependence on classical centralized systems would be questionable in terms of user privacy and transparency, as such models often act as“black boxes.”To address these issues, we propose and evaluate a conceptual framework, XFL-IoT, which combines Federated Learning (FL) for decentralized model training with data privacy and Explainable AI (XAI) to provide comprehensible explanations of the decisions to the user. The working of FL to train the machine-learning model for multi-class intrusion detection is through the Edge-IIoTset dataset. Post-hoc interpretability is provided via Shapley Additive explanations (SHAP) to explain model predictions on a per-instance basis. The results demonstrate that the combined XFL-IoT system can offer a high level of predictive performance when providing interpretable local and global explanations for model decisions. Ultimately, while this XFL-IoT paradigm offers an exciting framework for secure and transparent IoT security, it requires careful navigation and balancing the (latent) trade-offs that are embedded in the underlying tension between data privacy and model transparency, and ultimately model predictiveness. This work highlights the importance of resolving this trade-off for the successful implementation of the systems.

Özet (Çeviri)

There is an accelerating trend for governments to adopt digital transformation, which is clearly reflected in the growing presence of Internet of Things (IoT) deployments worldwide. These technologies have significantly enhanced the quality and performance of various services. However, this rapid increase in the adoption of the IoT has also raised significant security and privacy concerns, necessitating the development of new technological solutions to secure IoT systems. In these solutions, Intrusion Detection Systems (IDSs) play a crucial role. AI-powered IDSs have been developed as effective real-time intrusion detection and response solutions. However, the dependence on classical centralized systems would be questionable in terms of user privacy and transparency, as such models often act as“black boxes.”To address these issues, we propose and evaluate a conceptual framework, XFL-IoT, which combines Federated Learning (FL) for decentralized model training with data privacy and Explainable AI (XAI) to provide comprehensible explanations of the decisions to the user. The working of FL to train the machine-learning model for multi-class intrusion detection is through the Edge-IIoTset dataset. Post-hoc interpretability is provided via Shapley Additive explanations (SHAP) to explain model predictions on a per-instance basis. The results demonstrate that the combined XFL-IoT system can offer a high level of predictive performance when providing interpretable local and global explanations for model decisions. Ultimately, while this XFL-IoT paradigm offers an exciting framework for secure and transparent IoT security, it requires careful navigation and balancing the (latent) trade-offs that are embedded in the underlying tension between data privacy and model transparency, and ultimately model predictiveness. This work highlights the importance of resolving this trade-off for the successful implementation of the systems.

Benzer Tezler

  1. Explaining federated learning-based movie recommendations

    Federe öğrenme tabanlı film önerilerinin açıklanması

    EGE SOYARAR

    Yüksek Lisans

    İngilizce

    İngilizce

    2026

    Bilgisayar Mühendisliği Bilimleri-Bilgisayar ve KontrolÖzyeğin Üniversitesi

    Bilgisayar Mühendisliği Ana Bilim Dalı

    DOÇ. DR. REYHAN AYDOĞAN

  2. A new framework for decentralized social networks: Harnessing blockchain, deep learning, and natural language processing

    Merkezsiz sosyal ağlar için yeni bir çerçeve: Blok zinciri, derin öğrenme ve doğal dil işlemeyi kullanmak

    AMIR AL KADAH

    Yüksek Lisans

    İngilizce

    İngilizce

    2024

    Bilgisayar Mühendisliği Bilimleri-Bilgisayar ve KontrolSakarya Üniversitesi

    Yazılım Mühendisliği Ana Bilim Dalı

    DR. ÖĞR. ÜYESİ DENİZ BALTA

  3. Blokzincir tabanlı güvenilir ve açıklanabilir yapay zeka kullanan akıllı sağlık çatısının geliştirilmesi

    Development of smart health framework using blockchain-based trustworthy and explainable artificial intelligence

    KÜBRA ARSLANOĞLU

    Doktora

    Türkçe

    Türkçe

    2026

    Bilgisayar Mühendisliği Bilimleri-Bilgisayar ve KontrolFırat Üniversitesi

    Bilgisayar Mühendisliği Ana Bilim Dalı

    PROF. DR. MEHMET KARAKÖSE

  4. Tuş vuruşlarına dayalı kimlik doğrulama yöntemleri: Evrimi, zorlukları ve gelecek yönelimlerinin kapsamlı bir incelemesi

    Keystroke based authentication methods: A comprehensive review of evolution, challenges, and future directions

    NEBİL VURAL GÜNDOĞAN

    Yüksek Lisans

    Türkçe

    Türkçe

    2025

    Yönetim Bilişim SistemleriIşık Üniversitesi

    Bilgisayar Mühendisliği Ana Bilim Dalı

    DR. ÖĞR. ÜYESİ BARIŞ ÇELİKTAŞ

  5. The role of ai and machine learning in cloud intrusiondetection systems

    Bulut saldırı tespit sistemlerinde yapay zeka ve makineöğreniminin rolü

    ANAS CHEHADEH

    Yüksek Lisans

    İngilizce

    İngilizce

    2026

    Bilgisayar Mühendisliği Bilimleri-Bilgisayar ve KontrolBahçeşehir Üniversitesi

    Siber Güvenlik Ana Bilim Dalı

    DOÇ. DR. YÜCEL BATU SALMAN