Network Intrusion Detection System Using Convolutional Neural Networks for Network Traffic Analysis

Abstract

This paper presents a deep learning-based Network Intrusion Detection System (NIDS) using Convolutional Neural Networks (CNN) for effective network traffic analysis. Using the NSL-KDD dataset, we trained and evaluated various CNN architectures—ranging from single-layer to three-layer models.  The comparative analysis of single-layer, two-layer, and three-layer CNN models showed that increasing model depth does not necessarily improve performance and may, in fact, introduce marginal degradation due to overfitting or redundant feature representation. The two-layer CNN model achieved 99.90% accuracy, showing that deeper feature extraction did not significantly improve accuracy in this case, possibly because the features captured in the first layer were already highly separable. Interestingly, the three-layer CNN resulted in a slightly lower accuracy of 99.87%, which may be attributed to overfitting or redundancy in deeper feature representations when applied to a relatively well-structured dataset. The proposed system achieved outstanding classification accuracy, with the single-layer CNN reaching 99.92%, outperforming deeper models and several existing machine learning approaches. Experimental results show that shallow CNN architectures can effectively extract spatial features from structured traffic data, delivering superior performance with reduced computational complexity. A comparative analysis with recent state-of-the-art methods confirms the proposed model's efficiency and robustness, making it a viable and scalable solution for real-time intrusion detection in modern network environments.