Principal Component Analysis and Deep Learning-Based Traffic Anomaly Detection in WSN

Abstract

Wireless Sensor Networks (WSNs) have advanced quickly due to the fast expansion of wireless networks. Yet, because of their ease of use and versatility, security concerns have grown. This means that conducting research on intrusion protection in WSNs is now essential. Denial of Service (DoS) assaults are among the most common types of network attacks. They are dangerous because they take down the target network in order to accomplish their goal. Within WSNs, where devices function with limited resources, a denial-of-service attack has the potential to be disastrous. This research suggests a novel solution for WSNs, which are susceptible to assaults because to their devices' little storage capacity. To find abnormalities in DoS traffic within WSNs, the technique combines a Deep Convolutional Neural Network (DCNN) with Principal Component Analysis (PCA). By detecting and reducing the effects of DoS assaults, and by utilising the complementary capabilities of PCA and DCNN in this particular situation, the goal is to improve the security of WSNs. Compared with other traditional DL architectures, the proposed model has a more simplified structure and better feature extraction capabilities. This special combination gives it the power to quickly identify anomalous network activity in WSNs devices, especially those with limited storage. Because of its lightweight design, the suggested model addresses the inherent resource limits and guarantees optimal performance in the context of WSNs. A variety of assessment measures, such as confusion matrices, different classification metrics, and Receiver Operating Characteristic (ROC) curves, are used to verify the effectiveness of the suggested model. These metrics are used to evaluate the model's categorization performance in a rigorous manner. Extensive experimental comparisons reveal that the small size of the proposed model outperforms other popular models for anomalous traffic detection with regards to classification performance. This improved performance highlights how well the model works in WSNs to detect and handle anomalous network traffic