Deep-Learning-Based Network Intrusion Detection System for Edge Computing Environments

Edge computing environments host numerous devices with limited defense capabilities, rendering them vulnerable to malicious attacks. Developing network intrusion detection systems (NIDS) based on deep-learning techniques has emerged as a promising approach to address these vulnerabilities. However, the constrained computational power and storage capacity inherent to edge devices necessitate that such NIDS solutions be both lightweight and highly efficient in detection. To achieve this balance, this study employs Lasso regularization combined with K-fold cross-validation for feature selection, aiming to reduce input dimensionality while maintaining detection accuracy. Additionally, we propose a multilayer perceptron model incorporating a binary mask, trained using a straight-through estimator, to further enhance detection performance. The proposed NIDS was evaluated on two widely used network intrusion detection datasets. The results demonstrate strong classification performance, confirming the system's effectiveness while maintaining its essential lightweight characteristics.