AI-driven Throughput Prediction for Prioritization of Video Calling Traffic

The shift towards extensive video communication following the COVID-19 pandemic necessitates innovative traffic management solutions to maintain high-quality video calling experiences. Our previous work, Application Priority Engine (APE), focused on improving the video calling experience in the presence of background traffic. APE determines the trends in video calling bitrate using a simple Exponential Moving Average (EMA) and Simple Moving Average (SMA) comparison of the past 5 samples. It then uses the determined trends to control the background traffic. This paper addresses the shortcomings of this approach. Trend determination using past samples leads to a reactive traffic management system, which adjusts only after video quality deteriorates. To transform this into a proactive system, we explored using neural network models, specifically Bidirectional Long Short-Term Memory (Bi-LSTM) networks, for prediction of bitrate under Wi-Fi conditions. Our study collected a comprehensive dataset, encompassing bitrates, and various network parameters under different conditions using popular video calling applications. The proposed predictive traffic management can be extended to Internet of Things (IoT) networks, ensuring optimal bandwidth allocation and efficient operation.