Federated Learning and Convolutional Autoencoder for Robust Anomaly Detection in Agricultural IoT

Security is a vital challenge that must be addressed in the Agricultural Internet of Things (AIoT) to ensure enhanced productivity, quality, and resilience. Robust security measures are essential to protect sensitive data, prevent unauthorized access, and safeguard against cyberthreats, ultimately c...

Celý popis

Uloženo v:
Podrobná bibliografie
Vydáno v:2024 International Conference of the African Federation of Operational Research Societies (AFROS) s. 1 - 5
Hlavní autoři: Khacha, Amina, Saadouni, Rafika, Harbi, Yasmine, Gherbi, Chirihane, Aliouat, Zibouda, Harous, Saad
Médium: Konferenční příspěvek
Jazyk:angličtina
Vydáno: IEEE 03.11.2024
Témata:
Anomaly-based IDS
Autoencoders
Convolutional Autoencoder (CAE)
Deep learning
Distributed training
Federated learning
Independent and Identically Distributed (IID)
Internet of Things
Intrusion detection
Privacy-preserving
Productivity
Reliability
Resilience
Security
Training
On-line přístup:Získat plný text
Tagy: Přidat tag
Žádné tagy, Buďte první, kdo vytvoří štítek k tomuto záznamu!
Popis
Shrnutí:Security is a vital challenge that must be addressed in the Agricultural Internet of Things (AIoT) to ensure enhanced productivity, quality, and resilience. Robust security measures are essential to protect sensitive data, prevent unauthorized access, and safeguard against cyberthreats, ultimately contributing to agricultural operations' overall efficiency and reliability. One solution to improve security is the integration of intelligent Intrusion Detection Systems (IDS) using deep learning methods. However, traditional deep learning training methods require centralizing data and compromise client privacy. Federated learning (FL) addresses this challenge by enabling privacy-preserving and cooperative training of deep learning models among clients. This paper presents an anomaly-based IDS using a Convolutional Autoencoder (CAE) and FL. We assess our proposed solution using a new real and large dataset, CICIoT-2023. We evaluate our model with independent and identically distributed (IID) and non-IID data distributions to further enhance performance. Our model demonstrates superior performance compared to centralized training, achieving a recall of 99.99% and an F1score of 99.40%.
DOI:10.1109/AFROS62115.2024.11037027