Deep Learning Based Optimal Traffic Classification Model for Modern Wireless Networks
Putta Hemanth Kumar, Tuhina Samanta · 2022 IEEE 19th India Council International Conference (INDICON) · 2022
The Network traffic monitoring, identification, and classification is an active research area to identify the content, service, and the application which has generated this traffic. The network is classified into many different traffics, and encryption of that traffic is necessary for sensitive data privacy and information security. The encryption is done using TLS encryption, although formerly it was known as Secure Sockets Layer (SSL) in HTTPS traffic. So Identifying any information from encrypted traffic is a challenging task. The previous machine learning-based works on classification involve a high level of feature extraction from the traffic dataset to develop a model for traffic identification. Here the proposed deep learning model uses the SNI attribute. The model has Convolutional Neural Network layers and between these layers, we insert a small Multi-Layer Perceptron(MLP) and then connected it to the next Convolution layer. Unlike using fully connected layers, this model ends with Global Average Pooling which helps in reducing the overfitting and the complexity of the model. Thus resulting in the model with an accuracy of 99% for this dataset. On increasing the number of CNN layers, the model starts to overfit which in turn deteriorates the performance of the model.