Mitigating Cyber Threats in WSNs: An Enhanced DBN-Based Approach with Data Balancing via SMOTE-Tomek and Sparrow Search Optimization

Wireless sensor networks (WSNs), which include both fixed and moving sensors, are essential building blocks of cyberphysical systems. Each of these sensors senses, gathers, processes, and transmits data about their environment; they also self-organise and communicate via multi-hop links. Notwithstanding their importance, WSNs are vulnerable to damaging assaults that can interrupt operations in an instant. Current approaches to intrusion detection in WSNs face problems including false alarms, computational overhead, and low detection rates. High network correlation, data redundancy, and limited resources at sensor nodes are the root causes of these problems. But being linked to the IoT makes them susceptible to attackers. This study utilises a machine learning approach, specifically the Deep Belief Network (DBN) algorithm, to address the challenges of WSN intrusion detection. The algorithm is used to the input dataset to identify binary and multi-class attacks. Also, the input data is balanced using the SMOTE-TomekLink technique, which stands for Synthetic Minority Oversampling Technique Tomek Link. The balanced dataset that results from this blend’s synthesis of minority instances and elimination of Tomek linkages greatly improves WSN detection accuracy. to use the SMOTE-Tomek resampling method to address imbalanced WSN datasets, which helps to reduce the impact of overfitting and underfitting. To maximise the network’s precision, the Sparrow Search Algorithm (SSA) chooses the SMOTE’s sampling rate. to find the best model for WSN intrusion detection by doing a thorough evaluation utilising the WSN-DS dataset, which has 374,661 records. The outstanding performance of our model stands out as the key result of our research. Our proposal’s success in identifying and mitigating intrusions in WSNs is demonstrated by these data, which emphasise its efficiency and superiority in the context of WSN intrusion detection.