Deep learning approach for real-time video streaming traffic classification

Video streaming services such as Amazon Prime
Video, Netflix and YouTube, continue to be of enormous demands in everyday peoples’ lives. This enticed research in new mechanisms to provide a clear image of network usage and ensure better Quality of Service (QoS) for these applications. This paper proposes an accurate video streaming traffic classification model based on deep learning (DL). We first collected a set of video traffic data from a real network. Video streaming services such as Amazon Prime
Video, Netflix and YouTube, continue to be of enormous demands in everyday peoples’ lives. This enticed research in new mechanisms to provide a clear image of network usage and ensure better Quality of Service (QoS) for these applications. This paper proposes an accurate video streaming traffic classification model based on deep learning (DL). We first collected a set of video traffic data from a real network. Then, data was pre-processed to select the desired features for video traffic classification.

Based on the performance evaluation, the model produces an
overall accuracy of 99.3% when classifying video streaming
traffic using a multi-layer feedforward neural network. This
paper also evaluates the DL approach’s effectiveness compared
to the Gaussian Naive Bayes algorithm (GNB), one of the most
well-known machine learning techniques used in Internet traffic classification. The model is promising to be applied in a real-time scenario as it showed its ability to predict new unseen data with 98.4% overall accuracy.