A network traffic classification method based on comprehensive feature extraction and adaptive fusion networks

Abstract

Traffic classification techniques are essential for network management and security. As traffic types become more complex, relying solely on individual traffic features results in low accuracy, which is insufficient. To address this, an adaptive fusion network based on comprehensive features (CF-AFN) was proposed to improve network traffic classification accuracy. Specifically, a hybrid neural network extracts global traffic features, local traffic features, and statistical features. An adaptive fusion module then combines these features, effectively considering their heterogeneity. This approach efficiently leverages the strengths of different features to enhance the performance and reliability of network traffic classification. Experimental results on the public ISCX VPN-nonVPN2016 and CICIDS2017 datasets, using CF-AFN, demonstrate classification accuracies of up to 98.3 % and 97.12 %, respectively, outperforming eleven other traffic classification methods.