<![CDATA[Network intrusion detection remains a critical challenge in cybersecurity, particularly due to the increasing volume and complexity of network traffic. To address this issue, this study develops a deep learning framework that transforms tabular NSL-KDD data into image representations using the Lightweight Multi-feature Image Generator for Tabular Data (LM-IGTD). In addition, Homogeneous Noise Generation (HoNG) is applied to enrich feature diversity prior to processing. The transformed data are then classified using a Convolutional Neural Network (CNN) under a binary classification scheme to distinguish between normal and attack activities. Experimental results on the KDDTest+ dataset show that the proposed approach achieves an accuracy of 81.81%, an F1-score of 81.51%, and a ROC-AUC of 95.11%. The results indicate that LM-IGTD significantly contributes to improving the model’s ability to distinguish between classes, particularly in terms of ROC-AUC, while HoNG enhances classification performance in terms of accuracy and F1-score. However, a trade-off is observed between improved classification accuracy and the model’s probability ranking capability. Overall, these findings highlight that LM-IGTD provides an effective feature representation strategy, while HoNG offers a complementary contribution depending on the evaluation metric prioritized.]]>
Copyrights © 2026
