A federated learning-enabled energy-aware anomaly detection algorithm for secure big data analytics in IoT-based smart healthcare systems.

The rapid expansion of Internet of Things (IoT) devices in smart healthcare systems has led to the generation of large volumes of diverse medical data. This creates challenges in ensuring secure, scalable, and energy-efficient anomaly detection. Traditional centralized deep learning methods rely on continuously sending sensitive patient data to cloud servers, which increases communication overhead, consumes more energy, and raises privacy concerns. To overcome these limitations, this paper presents a Federated Learning-enabled Energy-Aware Anomaly Detection framework (FL-EAD) designed for IoT-based smart healthcare environments. The proposed approach allows distributed IoT devices to collaboratively train models while keeping patient data stored locally, sharing only model updates instead of raw data. An energy-aware client selection strategy is incorporated to determine device participation in each federated learning round based on factors such as residual energy and communication cost. This helps reduce unnecessary energy consumption. Furthermore, a hybrid deep learning model combining an autoencoder with an attention-based Long Short-Term Memory (LSTM) network is used to effectively capture both spatial and temporal patterns in healthcare data streams. The proposed framework is evaluated using a publicly available healthcare IoT dataset. Experimental results show that FL-EAD improves anomaly detection performance and overall system efficiency compared to traditional centralized methods and standard federated learning approaches. Notable improvements are observed in accuracy, F1-score, energy usage, communication overhead, and detection latency. Overall, the results suggest that the proposed framework offers a practical and privacy-preserving solution for scalable anomaly detection in next-generation IoT-enabled smart healthcare systems.