A NETWORK TRAFFIC ANOMALY CLASSIFICATION MODELBASEDON SELF ATTENTION MECHANISM AND CONVOLUTIONAL GATED RECURRENTUNIT

In this research, a network traffic anomaly classification model based on a Self-Attention Mechanism and Convolutional Gated Recurrent Unit (Conv-GRU) was proposed to improve the detection of malicious network activities. The rapid growth of internet services, cloud computing, and IoT devices has increased the complexity of network traffic, making traditional intrusion detection techniques less effective in identifying modern cyber threats. To address these challenges, the proposed model integrates convolutional neural networks for spatial feature extraction, GRU layers for capturing temporal dependencies in network traffic sequences, and a self-attention mechanism to focus on the most relevant traffic features. This hybrid architecture enables the system to effectively analyze complex network traffic patterns and improve anomaly classification performance.   The experimental evaluation using benchmark datasets demonstrates that the proposed model achieves higher detection accuracy and reduced false positive rates compared with conventional machine learning and basic deep learning approaches. The combination of CNN, GRU, and attention mechanisms allows the system to learn both spatial and temporal characteristics of network traffic data, thereby enhancing its ability to detect various cyber attacks such as DDoS, malware traffic, and unauthorized access attempts. These results indicate that the proposed approach can serve as an efficient and reliable solution for modern intrusion detection systems and network security monitoring applications.   Although the proposed model provides promising results, there are still opportunities for further improvement. In future work, the model can be extended to support real-time intrusion detection in large-scale network environments with higher data throughput. Additional optimization techniques may be applied to reduce computational complexity and improve processing speed. Furthermore, integrating advanced deep learning models such as transformer-based architectures or federated learning techniques could enhance detection performance and enable distributed security monitoring across multiple network nodes. Expanding the model to analyze encrypted traffic and incorporating adaptive learning mechanisms will also help improve its ability to respond to evolving cyber threats in dynamic network environments.  

[1]. Almuhanna, R., &Dardouri, S. (2025). A deep learning/machine learning approach for anomaly-based network intrusion detection. Frontiers in Artificial Intelligence, 8, 1625891. [2]. Alashjaee, A. M. (2025). Deep learning for network security: An Attention-CNN-LSTM model for accurate intrusion detection. Scientific Reports, 15, 21856. [3]. Wang, P., Chen, X., Shen, J., Xu, Z., Liang, F., & Du, Q. (2025). Abnormal traffic detection based on image recognition and attention-residual optimization. Frontiers in Communications and Networks. [4]. Liu, C., Liu, C., & Liu, C. (2025). An abnormal traffic detection method for chain information management system network based on convolutional neural network. Frontiers in Physics. [5]. Chen, X., Shen, J., Liang, F., & Du, Q. (2025). Network anomaly traffic detection model based on spatio-temporal attention feature fusion. Applied and Computational Engineering. [6]. Yang, Z., Jin, Y., Liu, J., Xu, X., Zhang, Y., & Ji, S. (2025). Research on cloud platform network traffic monitoring and anomaly detection system based on large language models. arXiv preprint. [7]. Peng, S., Wang, L., Shuai, W., Song, H., Zhou, J., Yu, S., & Xuan, Q. (2025). Hierarchical local-global feature learning for few-shot malicious traffic detection. arXiv preprint.   [8]. Guerra, L., Chapuis, T., Duc, G., Mozharovskyi, P., & Nguyen, V. (2025). Self-supervised learning of graph representations for network intrusion detection. arXiv preprint. [9]. Kouchaki, M., Zhang, M., Abdalla, A., Lan, G., Brinton, C., &Marojevic, V. (2024). Enhanced real-time threat detection in 5G networks using self-attention RNN autoencoder. arXiv preprint. [10]. Ji, C., Yu, H., & Dai, W. (2024). Network traffic anomaly detection based on spatiotemporal feature extraction and channel attention. Processes, 12(7), 1418.