An efficient data driven framework for intrusion detection in wireless sensor networks using deep learning

Wireless Sensor Networks (WSNs) are considered essential to distributed sensing in agricultural, health and industrial domains. Although WSNs have several advantages, they encounter profound cybersecurity threats owing to their processing capacities and small energy sources. In this research work, an intrusion detection framework based on deep learning is designed: a combination of Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs) with an adversarial-aware optimization model. The benchmark datasets of NSL-KDD, CICIDS2017, UNSW-NB15, and CTU-13 are analyzed in a number of ways based on structure, diversity, and deep learning requirements. We propose a compound objective to optimize all of these simultaneously, maximizing detection accuracy, minimizing adversarial vulnerability and ensuring model generalizability. Synthetic oversampling with SMOTE is employed to deal with this. Cross-dataset and intra-dataset experiments are implemented when testing the proposed framework, and it outperforms in terms of robustness and transferability. Our efforts are practical in terms of the deployment of a lightweight and resilient IDS that is suitable for WSN settings.