Federated Learning for Privacy-Preserving Intrusion Detection in IoT using Edge and Cloud Infrastructure

With the explosion of connected devices, IoT networks have become primeLATEX. targets for various forms of cyberattacks, compromising the confidentiality, integrity, and availability of systems. Traditional intrusion detection systems (IDS) based on centralized data collection are no longer suited for these distributed and heterogeneous environments. They pose significant privacy issues, latency concerns, and network overhead. In this context, I propose an innovative intrusion detection approach and new way based on Federated Learning (FL) integrated into a hybrid Edge-Cloud architecture. The goal is to design a privacy-preserving, distributed, and intelligent IDS that learns from local data on Edge nodes without ever transferring the data to a central server. Only model weights are aggregated at the Cloud level, thus reducing data leakage risks while conserving network resources. Our methodology involves deploying Deep Learning models (such as CNN or LSTM) trained locally on IoT data subsets, then federated via techniques like FedAvg. The system’s performance will be evaluated using standard metrics (accuracy, recall, F1-score), as well as latency, energy consumption, and bandwidth. Expected results aim to demonstrate that integrating Federated Learning into an Edge-Cloud infrastructure not only strengthens IoT network security but also ensures efficient, distributed, and privacy-compliant processing.