Inferring Video Quality in Live Streaming Flows Using Network Passive Metrics

Video streaming represents a substantial share of internet traffic, driven by the increasing demand for high-quality content and live broadcasts. This trend is particularly evident with the widespread adoption of HTTP-based adaptive bitrate streaming protocols, such as DASH and HLS. Internet Service Providers (ISPs) are often evaluated based on their customers' perceptions of premium services (e.g., video streaming), which are delivered over ISP networks by content providers like DAZN and Amazon Prime. While content providers have direct access to their customers' Quality of Experience (QoE), ISPs must infer this data from key performance indicators (KPIs) such as throughput, packet loss, and latency, especially given the growing prevalence of end-to-end encryption. This highlights the need for models capable of estimating QoE from passive metrics. In this research, we propose a methodology that leverages machine learning algorithms to infer video quality—one of the key QoE factors—using these passive metrics.