Blockchain-enabled federated learning framework with hybrid CNN-LSTM anomaly detection for secure edge IoT networks

Edge computing and Internet of Things (IoT) systems have introduced significant challenges in ensuring privacy, security, and trust in decentralized environments. This paper proposes FedChain-P2P, a novel federated learning framework that integrates blockchain-based trust management with deep learning–based anomaly detection for secure and scalable edge intelligence. Unlike conventional federated learning approaches that rely on centralized aggregation, the proposed framework leverages blockchain to establish decentralized trust among participating edge nodes, ensuring that only reliable devices contribute to model updates. The framework employs a hybrid CNN-LSTM model for anomaly detection to capture both spatial and temporal patterns in IoT telemetry data. This enables accurate detection of cyber threats in dynamic edge computing environments. To further enhance privacy protection, local differential privacy (LDP) is applied during model updates, preventing sensitive data leakage while maintaining model performance. The proposed framework is evaluated using two benchmark IoT cybersecurity datasets, Edge-IIoTset and TON_IoT, which simulate real-world industrial and IoT network attack scenarios. Experimental results demonstrate that FedChain-P2P achieves a detection accuracy of up to 97.2%, reduces communication overhead by 18.6%, and maintains strong privacy guarantees through differential privacy and encrypted aggregation mechanisms. Additionally, the blockchain-enabled trust management layer improves system resilience against adversarial and malicious participants. Overall, the proposed framework provides a secure, scalable, and privacy-preserving solution for intelligent edge computing, combining federated learning, blockchain-based trust management, and deep learning–based anomaly detection.