Scalable IoT anomaly detection
Rguibi Arjdal, Younes Asimi, Ahmed Asimi, Lahcen Oumouss · 2025
Securing the ever-growing Internet of Things (IoT) demands scalable anomaly detection that respects user privacy. Centralized approaches struggle with data confidentiality and resource limitations. This work investigates federated and split learning for privacy-preserving, resource-efficient anomaly detection in IoT networks. Federated learning enables collaborative model training on distributed devices, while split learning further distributes the workload by keeping model parts on each device. We address design considerations for these approaches, focusing on balancing communication overhead with robust privacy guarantees, all while adapting to evolving normal data patterns. Our research advances the use of federated and split learning for secure and scalable anomaly detection in resource-constrained IoT environments.