Network traffic analysis based on deep learning algorithms

Abstract

We propose a unified anomaly detection framework based on CNN-BiLSTM-Attention to address the low accuracy and high false positive rates of existing traffic-based network anomaly detection methods. The framework represents each network session as a 2D grayscale image, uses 2D-CNN for spatial feature extraction, BiLSTM for long-range temporal dependency capture, and an attention mechanism to highlight the most discriminative features. Experiments on the USTC-TFC-2016 and CICDDoS 2019 datasets show that this framework significantly outperforms state-of the-art methods in accuracy, detection rate, and false positive rate, demonstrating its effectiveness and generalization for network traffic classification.