Cloud-Native Hybrid Deep Learning Pipeline for Network Traffic Classification using Convolutional Recurrent Neural Network

Identifying encrypted traffic from emerging applications is important and challenging as traditional traffic classification approaches fail to achieve the desired level of accuracy. This necessitates the elaborate strategies to learn about the encrypted network flow. The research focuses on developing a cloud native hybrid Deep Learning (DL) pipeline for Network Traffic Classification (NTC). Initially, the ISCXVPN2016 dataset is considered that consists encrypted and non-encrypted traffic where the raw data is pre-processed with packet length normalization. The features are extracted applying Convolutional Neural Networks (CNNs) to define statistical flow features including inter-arrival time, packet size and flow duration. The data from network traffic is processed and classified using Convolutional Recurrent Neural Network (CRNN) with microservices executed in serverless containers. The hyper parameters in the DL models are then tuned using grid search and the random search algorithms. Experimental results showed that the proposed cloud native CRNN pipeline achieved accuracy of 98.67%, precision of 95.46%, and 94.32% when compared to the existing NTC models Stacked Autoencoders (SAE) based CNN (SAE-CNN) and Autoencoders based CNN (AE-CNN).