Wireless Sensor Networks Security Using SACCGAN‐RSA in a Novel Deep Learning–Based Intrusion Detection Framework for Enhancing

ABSTRACT Wireless sensor networks (WSNs) are essential to cyber–physical systems with their multiple hop, self‐organized arrangement of mobile sensor components. WSNs sense, aggregate, process, and transfer data across a geographical network area, ultimately forwarding it to the network's head. Nevertheless, existing intrusion detection systems (IDS) for WSNs face various limitations such as less detection rates, high computational costs, and significant false alarm rates due to resource constraints, redundancy, and data correlation. To address these challenges, a self‐attention–based cycle‐consistent generative adversarial network (SACCGAN), enhanced using reptile search algorithm for intrusion detection in WSN environments (SACCGAN‐RSA‐IDS‐WSN), is proposed. The data are collected through WSN‐DS dataset and supplied to the preprocessing stage. Preprocessing involves adjusted quick shift phase preserving dynamic range compression (AQS‐PDC) to address missing values and remove redundant data. The feature selection is done by Pelican optimization algorithm (POA) to identify ideal features. Utilizing the selected features, SACCGAN categorizes WSN data into normal and anomalous instances, including blackhole, grayhole, flooding, and scheduling attacks. However, SACCGAN lacks adaptation of optimization modes to ensure accurate WSN intrusion detection. Therefore, the reptile search algorithm (RSA) is introduced to enhance SACCGAN parameters effectively. The proposed SACCGAN‐RSA‐IDS‐WSN is executed in Python using WSN‐DS database. The SACCGAN‐RSA‐IDS‐WSN approach is examined under some metrics like recall, precision, F‐measure, specificity, accuracy, RoC, and computation time. The SACCGAN‐RSA‐IDS‐WSN method attains 28.59%, 31.43%, and 30.18% higher accuracy; 33.45%, 37.67%, and 27.78% higher ROC; 31.73%, 33.49%, and 29.15% lower computational time when compared with existing techniques.