A Convolutional Neural Network Approach to Improving Network Visibility

Abstract

Increasingly researchers are turning to machine learning techniques such as artificial neural networks to address communication network research questions. At the heart of each challenge is the need to classify packets and improve visibility. To date, multi-layer perceptron neural networks have been used to successfully identify individual packets. This work utilizes convolutional neural networks to classify packets after their conversion to an image matrix. To help address network challenges and aid in visualization, packets are combined into larger images to provide greater insight into a particular time span. Applications of this research can use the surrounding temporal area to gain insight into conversations, exchanges, losses and threats. We demonstrate the use of this technique to identify potential latency problems. This approach of using contemporary network traffic and convolutional neural networks has success rate for individual packets exceeding 99%. Larger images providing a broader view achieve the same high level of accuracy.