ETCNet: Encrypted Traffic Classification using Siamese Convolutional Networks

As more and more Internet traffic is encrypted, classifying their flows for the usage in application-aware networking (AAN) and application-network integration (ANI) becomes increasingly important and challenging. Traditional deep packet inspection approaches are no longer capable of identifying the encrypted packet streams, and hence new traffic classification methods based on machine learning have recently been explored by several researchers. One major challenge of using machine learning to classify encrypted traffic is lacking real datasets. Collecting Internet traffic may leak users' sensitive information, which prohibits the network community from sharing the datasets they collected. In this poster, we propose ETCNet which is a model based on Siamese convolutional network to solve this issue. Our evaluation for the ETCNet shows that it can achieve high accuracy by only using 40 flows of each application to train it.