Abstract: This paper proposes a network traffic classification method that combines feature generation and long short term memory (LSTM) model. This method analyzes and compares the classification performances of different feature generation methods using matrix multiplication feature generation method. The accuracy of original data and feature data on the classification problem is tested experimentally, and the results of convolutional neural network (CNN) and the proposed method are compared on network flow classification. The kernel function is used in the statistical feature, so that it can adapt to the LSTM input dimension and obtain better classification results. Experimental results on real network flow data show that the proposed method can achieve 93.9% accuracy in classification, and 99.2% in coarse grained classification task, and this performance is significantly better than that of existing methods.

Key words: traffic classification, feature generation, long short term memory (LSTM), fine classification