An Intrusion Detection System Based on Federated Deep Learning

Abstract

The development of networks and their technology has made it possible for our devices and data to be increasingly compromised, especially during data sharing and distribution. Intrusion detection systems (IDS) based on machine learning (ML) and deep learning (DL) approach can be a solution to this problem to deal with these attacks and threats. However, with many devices on our network, we may need to share data with the server for collection and training to build our model, which is still very risky for privacy and confidentiality. Federated learning (FL) is a suitable solution to this problem, which ensures that data is not shared with the server for training. Instead, it allows us to train locally and build our model. This model is then sent to the server to be aggregated and updated to build a new model, and so on. In this study, we propose a unified learning- based intrusion detection system using a neural network algorithm (CNN). This algorithm showed good results with the UNSW-NB15 dataset. In both the central and decentralized approaches, the results were close to better for current works, while ensuring data privacy and model security in the decentralized approach.