Explaining Deep Learning-Based Traffic Classification Using a Genetic Algorithm

Traffic classification is widely used in various network functions such as software-defined networking and network intrusion detection systems. Many traffic classification methods have been proposed for classifying encrypted traffic by utilizing a deep learning model without inspecting the packet payload. However, they have an important challenge in that the mechanism of deep learning is inexplicable. A malfunction of the deep learning model may occur if the training dataset includes malicious or erroneous data. Explainable artificial intelligence (XAI) can give some insight for improving the deep learning model by explaining the cause of the malfunction. In this paper, we propose a method for explaining the working mechanism of deep-learning-based traffic classification as a method of XAI based on a genetic algorithm. We describe the mechanism of the deep-learning-based traffic classifier by quantifying the importance of each feature. In addition, we leverage the genetic algorithm to generate a feature selection mask that selects important features in the entire feature set. To demonstrate the proposed explanation method, we implemented a deep-learning-based traffic classifier with an accuracy of approximately 97.24%. In addition, we present the importance of each feature derived from the proposed explanation method by defining the dominance rate.