Network Flow Classification and Volume Prediction using Novel Ensemble Deep Learning Architectures in the Era of the Internet of Things (IoT)

Abstract

Network flow and volume optimization have become a challenging task with the rapid growth of IoT devices. Fortunately, network flow classification and network volume prediction techniques are powerful tools to allocate resources efficiently in a given network. Current machine learning models have proven useful since they partition and direct traffic to extremely diverse sets of devices and services. Current machine/deep learning techniques are geared towards specific network configurations and suffer from suboptimal accuracy or lengthy training methodologies. In this paper, we propose a family of novel ensemble techniques for short-term network volume predictions and flow classifications. We explore each architecture and the characteristics of each model to achieve high efficiency. Experimental results on three real network traffic datasets show that the proposed family of deep ensemble models using Long Short-Term Memory (LSTM) and Convolutional LSTM neural network (CNN_LSTM) are best-in-class models for short-term flow classification and volume predictions.