Classification of darknet activities using neural networks

It is very important to characterize and analyze the network before being exposed to threats. In this study, darknet network traffic analysis was carried out and the darknet network was determined and characterized by examining the 2 layer artificial neural network models. In the first layer, it distinguishes whether the data is benign or darknet traffic, and in the second layer, it is determined which of the categories of Browsing, P2P, Chat, Email, Transfer, Audio/Video Stream and VOIP generated by the traffic. Experiments were made with RNN, LSTM and MLP algorithm models. New data sets were produced with GAN and used as training data. LSTM and MLP algorithms are reconstructed as both multi category and binary category. The feature selection algorithm has been applied in the MLP model. CICDarknet2020 dataset was used. According to the model accuracy values, RNN, 0.98 success was achieved in the detection of darknet traffic, and 0.86 in the second layer. In the LSTM model, values of 0.99 and 0.71 were obtained. Separately modeled categories were obtained as 0.92. In the MLP model, accuracy values of 0.99 and 0.78 were observed. The close outputs were obtained with the feature selection algorithm model. In the binary category model, 96% accuracy was achieved