Network traffic analysis based on deep learning algorithms

Abstract

This paper introduces a novel hierarchical model integrating 2D-CNN and Bi-LSTM-Attention for improved network traffic analysis. We standardized the USTC-TFC-2016 and CICDDoS 2019 datasets into 40 × 40 grayscale images, extracting spatial features with 2D-CNN and temporal dependencies with Bi-LSTMAttention. Our model demonstrated superior accuracy and robustness over conventional approaches, with capabilities for real-time malicious traffic detection.