Transfer Learning for Encrypted Malicious Traffic Detection Based on Efficientnet

With the development of data encryption technology, encrypted traffic has shown an explosive growth trend. More and more malicious network services rely on encryption to evade detection, which brings huge challenges to traditional rule-based traffic classification methods. In recent years, artificial intelligence technology provides feasible ideas to solve this problem. However, conventional machine learning depends on expert experience for manual feature extraction, and the training of deep learning models requires large number of high-quality labeled data, which makes the research of encrypted traffic detection hard to conduct. To address these issues, we propose a transfer learning method based on Efficientnet to detect the encrypted malicious traffic. Efficientnet-B0, a pre-trained model based on the Imagenet dataset, is transferred to an encrypted traffic dataset, extracting features automatically from the original traffic data without expert experience. The experimental results show that our method can achieve 100% detection accuracy and recall rate. Furthermore, our method can also achieve high detection performance with a small amount of training samples.