A HYBRID DEEP LEARNING FRAMEWORK FOR ENHANCED THREAT DETECTION IN WIRELESS SENSOR NETWORKS

Abstract

Wireless Sensor Networks (WSN) have become a crucial technology in various areas. They are adopted across a wide range of distributed and data-driven applications. Due to the decentralized and resource-constrained nature of WSN, these systems are vulnerable to various security threats. These systems are susceptible to multiple forms of malicious activities that disrupt communication and degrade network performance. Traditional methods are unable to recognize the intricate temporal and spatial patterns found in network traffic data. So, these systems frequently fail to detect advanced threats. For effective intrusion detection, this paper presents a hybrid deep learning framework that combines Convolutional Neural Networks (CNN), Long Short-Term Memory (LSTM), and an attention mechanism. High-level spatial features are extracted from network traffic via the CNN component. Additionally, temporal dependencies between sequential data are captured by the LSTM. By giving important time steps adaptive priority, the attention mechanism improves the model's detection accuracy for dynamic attack behaviors. The WSN-DS dataset is used in this experiment. The model obtains an overall accuracy of 98.75% along with high performance in other metrics across several attack classes. The research results show that the proposed design performs noticeably better than both independent Deep Learning (DL) models and traditional Machine Learning (ML) techniques.