Convolutional Neural Networks for Real-Time Intrusion Detection in Network Security Architectures

Intrusion detection remains one of the most critical challenges in safeguarding modern network infrastructures, particularly as cyberattacks continue to evolve in scale and sophistication. Conventional approaches, often reliant on static signatures or shallow classifiers, struggle to provide timely and accurate detection in dynamic traffic environments. This research introduces a novel convolutional neural network (CNN)-based model designed for real-time intrusion detection, highlighting its uniqueness in capturing temporal burstiness, protocol behaviors, and anomalous traffic distributions within large-scale packet data. The analyzed dataset consists of over one million traffic records, encompassingboth normal and malicious instances, although it ishighly imbalanced in favor of benign flows. Exploratory analysis revealed clear patterns in packet sizes, protocol-port usage, and temporal spikes of malicious activity, which were effectively leveraged to strengthen feature learning. The proposed model integrates one-dimensional convolutions, dilated layers, and attention mechanisms to extract hierarchical patterns and improve minority-class detection. Baseline comparisons with gradient boosting demonstrated meaningful predictive power, while the CNN achieved superior results when evaluated through confusion matrices, ROC curves, and precision-recall curves. The novelty of this research lies in embedding temporal and contextual features into the CNN framework, enabling enhanced classification accuracy even under severe imbalance conditions. This quantitative study demonstrates that the proposed CNN architecture provides a scalable, adaptive, and high-performing solution for intrusion detection, making it particularly well-suited to the demands of real-time and 6G-ready network environments.