A Novel Distributed Learning Approach for Detecting Covert Communications in Wireless Sensor Networks

ABSTRACT Covert communication, a critical cybersecurity threat, involves the use of concealed or subtle methods to transmit information within a network without detection. This can include disguising communication within normal network traffic or using steganography to evade traditional monitoring techniques. This paper presents a novel approach for detecting covert communication within Wireless Sensor Networks (WSNs) using an enhanced anomaly detection model called ADSS‐GAN‐DL. The proposed model integrates Generative Adversarial Networks (GANs), Adaptive Synthetic Sampling (ADASYN), and a Distributed Learning (DL) mechanism to effectively identify subtle deviations in network traffic that are characteristic of covert communication. Covert communication, often concealed within normal traffic, poses significant challenges for traditional anomaly detection models due to its subtle nature. The ADSS‐GAN‐DL model addresses this by leveraging GANs to generate realistic network traffic and training a discriminator to detect slight anomalies. ADASYN handles class imbalance, ensuring better detection of rare covert communication instances, while the Distributed Learning mechanism enables collaborative data processing across multiple network nodes, enhancing scalability and efficiency. Experimental results demonstrate a detection accuracy of 95.8%, with improvements in precision (96.2%) and recall (94.7%) compared to traditional models. These results validate the model's capability in detecting covert communication, making it a robust solution for network security in distributed environments.