Deep Learning-Based Intrusion Detection for Wireless Sensor Networks Employing Dimensionality Reduction and Data Balancing Techniques

Abstract

– Wireless Sensor Networks (WSNs) have become imperative in the areas of healthcare, industrial automation, military surveillance, environmental monitoring and smart city applications, hence the need to develop solid intrusion detection systems (IDSs) that will guarantee security of the network. Nonetheless, the constraint of WSNs in terms of resources and the extreme imbalance of classes within intrusion detection data prove to be an overwhelming impediment to the capacity of deep learning (DL) models to identify minority attack classes with high accuracy, leading to poor detection outcomes. To overcome these challenges, this paper presents a systematic DL-based intrusion detection framework comprising three advanced hybrid data balancing methods (KMeans-SMOTE, SMOTE-ENN, and SMOTE-Tomek Link) and nine DL architectures using the WSN-DS dataset. The recommended framework decreases classification bias, learns minority attack