A Hybrid Cloud–Edge Lightweight Deep Learning Framework for Real-Time Anomaly Detection in IoT-Based Wireless Sensor Networks

Abstract

The rapid expansion of the Internet of Things (IoT) has led to the deployment of billions of wireless sensor nodes across various critical infrastructures. While these sensor networks provide valuable real-time monitoring capabilities, their inherent resource limitations make them vulnerable to various sophisticated cyber-attacks. Traditional security solutions often rely on centralized cloud computing for processing intrusion detection tasks. However, this centralized approach introduces significant communication delays and consumes substantial network bandwidth, making it unsuitable for time-sensitive applications. To address these challenges, this paper proposes a hybrid cloud-edge framework for real-time anomaly detection in wireless sensor networks. The proposed framework distributes computational tasks between the edge and the cloud layers to achieve high detection accuracy while maintaining low response time. At the edge layer, a lightweight Random Forest model is deployed to perform rapid initial filtering of network traffic, effectively identifying common attack patterns locally. Anomalies that require deeper investigation are transmitted to the cloud layer, where a robust CNN-LSTM model performs exhaustive forensic analysis to detect complex and multi-stage intrusions. Furthermore, the communication between the edge and the cloud is secured using a hybrid AES-ECC encryption scheme to ensure data confidentiality and integrity. Experimental results using the NSL-KDD and CICIDS2017 datasets demonstrate that the proposed framework achieves an F1-score of 98.7 percent and a classification accuracy of 99.1 percent. The hybrid architecture reduces wide-area network traffic by 65 percent and provides a hardware-level latency reduction of 92 percent compared to standalone cloud-based systems