Multi-Scale Convolutional Neural Networks optimized by elite strategy dung beetle optimization algorithm for encrypted traffic classification

The rapid development of the Internet has resulted in a wide range of traffic types. Encrypted traffic was once considered the most secure option for online browsing and conducting business. However, with advancements in network technology, many network threats exist within encrypted channels, such as VPN and Tor, which makes encrypted traffic identification crucial. In this work, we design an improved Multi-Scale Convolutional Neural Network (MSCNN) model whose hyperparameters are optimized and adjusted by the elite strategy dung beetle optimization (ESDBO) algorithm by improving the initialization and population update strategy of the regular DBO algorithm. By introducing chaotic sequence initialization and elite strategy, the convergence velocity of the algorithm is increased, and the defect of the algorithm easily falling into local optima is improved, which makes the generated hyperparameters more suitable for the encrypted traffic identification, further enhancing the model's classification performance. We evaluate the performance of our model, using the ISCXVPN2016 dataset, ISCXTor2016 dataset and Cross-Platforms (Android and iOS) datasets for multi- class classification, respectively. The experimental results demonstrate that our model can effectively classify encrypted traffic with an overall accuracy of 86.77% in the ISCXVPN2016 dataset, which surpasses the comparative method by 4.64%. And it also achieves the accuracy of 85.64% in the relatively more imbalanced ISCXTor2016 dataset. Furthermore, our proposed ESDBO-MSCNN also achieves the best performance on Cross-Platforms (Android and iOS) datasets.