Abstract
This research presents a novel Explainable Adversarial Defence Framework to mitigate Distributed Denial of Service (DDoS) attacks in IoT edge networks while preserving data privacy. The framework employs a hybrid deep learning model utilising Convulational Neural Network (CNN) and Bi-Directional Long Short-Term Memory (BiLSTM) networks, trained via Federated Learning. It uses SHAP and LIME Explainable AI techniques to enhance decision transparency. Evaluation on the CICIIoT2025 dataset shows impressive accuracy of 93.84%, and a precision of 99.6%, demonstrating that the Federated Learning approach maintains centralised performance of 99.6% while improving privacy. Cross dataset validation reveals moderate robustness to domain shifts of 89.2% accuracy, with insights for proactive defence, making it suitable for decentralised IoT security systems. Key contributions include the first use of CNNBiLSTM-Attention for federated IoT DDoS detection together with the integration of SHAP and LIME for explainable security decisions, and the comprehensive evaluation across multiple metrics such as latency and communication overhead.