HybridNet-TC: A CNN-LSTM Architecture with Attention Mechanism for Real-Time Network Traffic Classification

The rapid proliferation of network traffic driven by cloud computing, IoT deployments, and encrypted communication protocols has rendered traditional signature-based classification methods inadequate for modern network environments. In this paper, we propose HybridNet-TC, a novel deep learning framework that synergistically integrates one-dimensional Convolutional Neural Networks (۱D-CNN) for spatial feature extraction with Long Short-Term Memory (LSTM) networks for temporal dependency modeling, augmented by a channel-wise attention mechanism to prioritize discriminative traffic features. The proposed architecture is evaluated on three benchmark datasets CICIDS-۲۰۱۷, UNSW-NB۱۵, and ISCX-۲۰۱۲ achieving classification accuracies of ۹۹.۴۱%, ۹۸.۸۷%, and ۹۸.۶۳%, respectively. Comparative analysis against state-of-the-art baselines, including standalone LSTM, Random Forest, SVM, and XGBoost, demonstrates that HybridNet-TC consistently outperforms existing approaches across all evaluation metrics while maintaining a real-time inference latency of ۳.۲ milliseconds per batch. Ablation studies confirm the complementary contribution of each architectural component. These results establish HybridNet-TC as a scalable and practical solution for real-time traffic classification in enterprise and carrier-grade network environments

کلیدواژه ها:

Network traffic classification ، deep learning ، convolutional neural network ، LSTM ، attention mechanism ، network security