Network Traffic Classification Method Based on Improved Capsule Neural Network

Abstract

Convolution neural network (CNN) has achieved great performance in network traffic classification problem. However, it needs large-scale training set to achieve better classification performance while decreases the accuracy result in the case of the small dataset. To solve this problem, this paper proposed a traffic classification algorithm based on the improved capsule neural network (CapsNet), which replaces the scalar feature output of CNN with vector output, and replaces the max-pooling with consistent routing, and disturbs some values in the capsule to reconstruct images. The experimental results show that the proposed method have a robust performance compared with the existing CNN method when processing small data sets, which corresponds the fact that the grayscale images output by the reconstruction module can make the classification results easier to understand.