Adaptive Deep Learning-Driven Framework for Malicious Node Detection in Wireless Sensor Networks

Wireless Sensor Networks (WSNs) are essential for many applications, such as military surveillance, healthcare, and environmental monitoring. However, because of their resource limitations and dynamic nature, they are susceptible to malicious node attacks, which jeopardize data integrity and network performance. In order to detect malicious nodes in dynamic WSN environments, this article proposes a hybrid Deep Learning-based architecture that combines Long Short-Term Memory (LSTM) networks with Deep Convolutional Neural Networks (DCNN) using weighted score level fusion. The suggested model with balanced dataset using proposed SMOTE-k algorithm correctly classifies nodes based on behavioral patterns and network dynamics by utilizing the feature extraction powers of DCNN in conjunction with the temporal sequence analysis of LSTM. To improve the model's performance and lessen the effect of unbalanced datasets, data balancing techniques were used. Important measures like accuracy, precision, recall, and F1-score—which were obtained from the confusion matrix—were used to assess the model's performance. A comparison with the previous work [18] shows that the hybrid DCNN-LSTM model provides a reliable and effective way to enhance security in WSNs, achieving superior accuracy (99.80%). The article outlines possible avenues for further research and emphasizes how well deep learning techniques work to overcome WSN security issues.