A Survey of Machine Learning Techniques for Video Quality Prediction from Quality of Delivery Metrics

A growing number of video streaming networks are incorporating machine learning (ML) applications. The growth services places enormous pressure on network and content providers who need to proactively maintain high levels quality. ML has been applied predict the quality streams. Quality delivery (QoD) measurements, which capture end-to-end performances services, have leveraged in prediction. drive for encryption, privacy digital rights management, brought about a lack visibility operators desire insights from metrics. In response, numerous solutions proposed tackle challenge prediction QoD-derived This survey provides review studies that focus techniques predicting QoD metrics services. context we metrics, not tied particular type service. Unlike previous reviews area, this contribution considers papers published between 2016 2021. Approaches grouped under following headings: (1) impairments, (2) encrypted traffic, (3) HAS applications, (4) SDN (5) wireless settings, (6) WebRTC Throughout survey, some research challenges directions area discussed, including over deep learning; adaptive improved delivery; computational cost interpretability; self-healing failure recovery. findings reveal traditional algorithms most widely adopted models solving problems. family lot potential because they well understood, easy deploy, lower requirements than techniques.