Dynamic and Explainable Federated Learning for IoT Anomaly Detection: A Comparative Study with Centralized Machine Learning Models

Jarjis A., BECERİKLİ Y.

4th International Conference on Innovations in Computing Research, ICR 2025, London, İngiltere, 25 - 27 Ağustos 2025, cilt.1487 LNNS, ss.211-227, (Tam Metin Bildiri) identifier

  • Yayın Türü: Bildiri / Tam Metin Bildiri
  • Cilt numarası: 1487 LNNS
  • Doi Numarası: 10.1007/978-3-031-95652-2_19
  • Basıldığı Şehir: London
  • Basıldığı Ülke: İngiltere
  • Sayfa Sayıları: ss.211-227
  • Anahtar Kelimeler: Explainable AI, Federated Learning, IoT Anomaly Detection, Lime, Machine Learning, SHAP
  • Kocaeli Üniversitesi Adresli: Evet

Özet

This paper presents a comparative analysis of centralized and federated learning (FL) methods for Internet of Things (IoT) anomaly detection, emphasizing model transparency explainability. We simulate a five-client FL setup using real-world IoT datasets to assess how model interpretation changes between local and global aggregation. Our results reveal that while centralized models capture global feature importance effectively, FL introduces inconsistencies due to data heterogeneity and local overfitting. Analysis shows feature drift across clients, while highlighting local decision logic variations. The findings underline the trade-off between privacy-preserving collaborative learning and model interpretability, which is essential for securing IoT networks.