CSCNN: Cost-Sensitive Convolutional Neural Network for Encrypted Traffic Classification

By the rapid development of the Internet and online applications, traffic classification not only has changed to an interesting topic in the field of computer networks but also plays a key role in cyber-security and network management. Although numerous studies have been conducted in recent years, encrypted traffic classification still remains a major challenge and unbalanced data is known as one of the most important problems in this field. Even though previous researches have focused on dealing with the class imbalance problem in the pre-processing step via machine learning and specifically deep learning methods, they are still confronted with some restrictions. To this end, a new traffic classification method is presented in this paper that aims to deal with the problem of unbalanced data along the training process. The proposed method utilized a Cost-Sensitive Convolution Neural Network (CSCNN) where a cost matrix was employed to assign a cost to each misclassification based on the distribution of each class. These costs were then utilized during the training process to increase the final classification accuracy. Various experiments were carried out to explore the performance of the proposed method for the tasks of traffic classification, traffic description, and application identification.‌ According to the obtained results, CSCNN achieved higher efficiency compared to both machine learning and deep learning based methods on the ISCX VPN-nonVPN dataset.