LSTM-JSO framework for privacy preserving adaptive intrusion detection in federated IoT networks

Abstract

The rapid proliferation of Internet of Things (IoT) systems has created unprecedented opportunities for smart environments and introduced critical cybersecurity vulnerabilities. Existing intrusion detection systems often fail to address sophisticated distributed attacks, such as botnets and Distributed Denial of Service (DDoS), due to their reliance on centralized data and static configurations. This research introduces a novel framework that enhances intrusion detection in IoT systems through a decentralized approach. The framework dynamically optimizes hyperparameters using the Joint Strategy Optimization (JSO) algorithm, ensuring efficient model adaptation across heterogeneous IoT networks. A Long Short-Term Memory (LSTM)-based architecture accurately captures complex intrusion patterns. Federated learning is leveraged to enable collaborative model training across multiple IoT networks while maintaining data privacy and reducing vulnerability to data poisoning. The proposed model's robustness and adaptability were validated using Transformer, bi-LSTM, and GRU architectures. This framework demonstrates the potential to address evolving cyber threats and provide robust security solutions for IoT ecosystems.