Federated Learning With Sailfish-Optimized Ensemble Models for Anomaly Detection in IoT Edge Computing Environment

Abstract

The Internet of Things (IoT) has transformed cyber-physical systems by enabling seamless connectivity and automation. However, IoT devices face resource constraints, making anomaly detection challenging. Traditional centralized approaches suffer from computational inefficiencies, increased latency, and privacy concerns, making them unsuitable for real-time anomaly detection in distributed IoT environments. To address these challenges, this paper proposes a privacy-preserving anomaly detection framework that integrates Federated Learning (FL) with an optimized Isolation Forest model. FL enables decentralized training on IoT devices, reducing the risk of data breaches. However, anomaly detection performance is often hindered by suboptimal parameter selection. To overcome this, the Sailfish Optimization Algorithm (SFO) is incorporated to fine-tune the Isolation Forest model’s parameters dynamically, balancing exploration and exploitation. This optimization enhances accuracy while maintaining data confidentiality. Additionally, the framework is evaluated against leading FL-based and traditional anomaly detection models, including Local Outlier Factor (LOF), Generative Adversaria (GAN), and Variational autoencoder (VAE), demonstrating superior performance in recall and F1-score. Extensive experiments on benchmark datasets confirm that the proposed method achieves higher anomaly detection efficiency with a lower error rate than existing methods. The results establish this framework as a scalable, privacy-preserving, and computationally efficient solution for anomaly detection in IoT edge environments, addressing critical limitations in security, latency, and data privacy in real-world applications.