Federated Deep Learning for Cybersecurity and Intrusion Detection in Decentralized Networks

The advancements in modern technologies demonstrate a growing trend in cyber attacks with complex patterns. These attacks can take various complex forms due to the diverse nature of data. To tackle this growing threat, ensuring data diversity is essential. However, contemporary works utilizing centralized model training face the challenge of maintaining data privacy. In this study, a federated deep learning based privacy-preserving architecture for Network Intrusion Detection is presented. This approach combines CNN and LSTM architecture within a Federated Learning (FL) framework. The framework was evaluated on a benchmark dataset CIC-IDS2017, encompassing seven classes - DoS, Portscan, Infiltration, Brute-force, Bot, Web Attack, and Benign. This framework demonstrates robustness in accurately detecting all the attack types with an accuracy of over 99.36%. This work will contribute to the cyber security field by addressing critical gaps in existing research and providing a robust solution for intrusion detection in a privacy-preserving manner.