Privacy-Preserving in Agricultural IoT: Intrusion Detection Using Federated Learning and CNN
Amina Khacha, Rafika Saadouni, Yasmine Harbi, Zibouda Aliouat, Chirihane Gherbi, Saad Harous · 2024
The Agricultural Internet of Things (AIoT) network integrates agriculture with IoT to enhance productivity and improve quality. This infrastructure connects many devices, facilitating extensive information exchange within the network. However, this increased data exchange also heightens the network's susceptibility to attacks and vulnerabilities, posing risks and potentially causing damage to the network infrastructure. Intrusion Detection Systems (IDS) are effective solutions for detecting attacks and preserving productivity. Many existing approaches in the literature combine deep learning with IDS to create robust anomaly-based IDS. However, these solutions often encounter challenges related to data collection, which compromises user privacy by centralizing data from various entities into a single entity. Our paper proposes a privacy-preserving IDS based on federated learning techniques and Convolutional neural networks (CNN) to develop a robust IDS with significant performance. We evaluate our experiments using a new dataset, Edge-IIoT, which consists of real IoT traffic across binary, 6-class, and 15-class scenarios with 10 and 15 clients. The results demonstrate the performance of our model across all scenarios, achieving 100% and 99.99% accuracy in binary classification and both 6-class and 15-class scenarios, respectively.