Lightweight Intrusion Detection Model Based on Convolutional Neural Network

The swift increase in cyber threats, projected to lead to global economic losses exceeding $10.5 trillion by 2025, underscores the urgent need for effective intrusion detection systems (IDS). Although deep learning-based IDS exhibit remarkable detection capabilities, their considerable computational complexity and large parameter scales pose significant challenges for practical implementation in resource-constrained environments. To address this challenge, we propose a lightweight convolutional neural network (L-CNN) architecture integrated into a comprehensive platform that utilizes the React and Django frameworks. The L-CNN reduces parameter complexity through a hierarchical structure comprising cascaded convolution-pooling-activation modules, further enhanced by dropout regularization and focal loss to improve minority class representation. Evaluated on the KDD Cup 99 dataset, our model achieves an accuracy of 97.15% across 39 attack categories, surpassing existing models while significantly minimizing computational overhead. This study demonstrates the feasibility of implementing deep learning-based IDS in edge computing environments while preserving detection accuracy, thereby offering a scalable solution for practical cybersecurity applications.