Encrypted Traffic Classification of Decentralized Applications on Ethereum using Feature Fusion

With the prevalence of blockchain, more and more Decentralized Applications (DApps) are deployed on Ethereum to achieve the goal of communicating without supervision. Users habits may be leaked while these applications adopt SSL/TLS to encrypt their transmission data. Encrypted protocol and the same blockchain platform bring challenges to the traffic classification of DApps. Existing encrypted traffic classification methods suffer from low accuracy in the situation of DApps. In this paper, we design an efficient method to fuse features of different dimensions for DApp fingerprinting. We firstly analyze the reason why existing methods do not perform well before proposing to merge features of different dimensions. Then we fuse these features by a kernel function and propose a fusion feature selection method to select appropriate features to fuse. Applying features that have been fused to the machine learning algorithm can construct a strong classifier. The experiment results show that the accuracy of our method can reach more than 90%, which performs better than state-of-the-art classification approaches.