A New Imbalanced Encrypted Traffic Classification Model Based on CBAM and Re-Weighted Loss Function

The accurate classification of traffic data is challenging for network management and security, especially in imbalanced situations. The limitation of the existing convolutional neural networks is that they have problems such as overfitting, instability, and poor generalization when used to classify imbalanced datasets. In this paper, we propose a new imbalanced encrypted traffic classification model. The proposed model is based on the improved convolutional block attention module (CBAM) and re-weighted cross-entropy focal loss (CEFL) function. The model exploits the redefined imbalance degree to construct a weight function, which is used to reassign the weights of the categories. The improved CBAM based on the redefined imbalance degree can make the model pay more attention to the characteristics of the minority samples, and increase the representation ability of these samples. The re-weighted CEFL loss function can be used to expand the effective loss gap between minority and majority samples. The method is validated on the public ISCX Tor 2016 dataset. The experimental results show that the performance of the new classification model is better than the baseline methods, and the proposed method can remarkably push the precision of the minority categories to 93.28% (14.63%↑), recall to 91.71% (16.98%↑), and F1 score to 92.49% (16.23%↑).