FLAME-X: Federated Learning with Adaptive Model Ensembles and Explainable AI for Real-Time DDoS Detection in IoT Networks

We propose FLAMEX, a hierarchical federated learning (FL) framework that integrates adaptive model ensembles and explainable AI (XAI) for DDoS detection in IoT networks. FLAMEX uses lightweight autoencoders for anomaly detection and XGBoost for attack classification at local IoT devices. Edge nodes aggregate latent features, while a central server applies DBSCAN clustering to handle non-IID data and personalize global models. SHAP explains feature contributions. Evaluated on the N-BaIoT dataset, FLAMEX achieves 99.2% accuracy for known attacks and an 0.8% false positive rate for zero-day threats. It reduces communication overhead by 52% compared to FedAvg via quantization and sparsification while ensuring privacy through differential privacy mechanisms.