Convolutional neural network-based identification of malicious traffic for TLS encryption

By converting network traffic into images and using convolutional network prediction methods, the accuracy of malicious network traffic can be effectively improved. As TLS traffic encryption is used in more and more network applications, while protecting security and privacy, network attackers can achieve the purpose of evading detection by encrypting malicious traffic. The traditional method of using a firewall and anti-virus software to decrypt the encryption key to access the encrypted data is not only inefficient and destroys the original purpose of encryption to protect privacy, but also when the key cannot be decrypted, the traditional method The decryption capability of the traditional method is insufficient. In this paper, we propose a method to transform network traffic into network fingerprint images without decrypting the encrypted data, and then feed the images into a convolutional neural network for prediction to achieve the classification of malicious traffic. The experimental results show that the method of identifying network fingerprint images by using CNN has high accuracy and prediction capability.