• Privacy
• IoT
• IDS
• Federated Learning
• Cybersecurity
• Smart Buildings

The integration of IoT and networking technologies in smart buildings has led to an increased need for security measures to protect against threats and vulnerabilities that could compromise system functionalities, data privacy, and even safety. This work highlights the importance of hardening against security attacks using Artificial Intelligence (AI) solutions, while maintaining confidentiality of private data. The research explores the collaborative training of Machine Learning (ML) models across multiple smart buildings, using Federated Learning (FL) techniques to distribute the model creation without directly sharing sensitive data, thereby preserving privacy by design. First, a comprehensive review of the current landscape in AI security, FL, Explainable AI (XAI), and security of 5G and 6G networks was conducted. This research led to the development of two convolutional neural network (CNN) models for detecting network and IoT attacks, using the public ToN-IoT dataset. Following a data engineering process that included sampling, feature extraction and transformation of data into images, a centralised model was first elaborated. The model was then successfully decentralised using TensorFlow Federated, achieving comparable performance to the centralised version. To address the unique vulnerabilities of federated learning, a secure and robust aggregation method was introduced, making the system resistant to poisoning attacks, from up to 20% of the participating clients.