Harnessing Crowdsourcing and Machine Learning for Mobile Services Performance Analysis

Data gathering from user terminals or specialized collection devices is required for crowdsourcing the evaluation of quality of experience (QoE) on mobile networks. With this strategy, mobile operators and academics may more affordably address large-scale issues, enhance resource allocation, and supply networks. Based on this strategy, we present here a taxonomy of dataset generation though mobile signals measurements, and also a comprehensive study utilizing crowdsourced data from user terminals to analyze the QoE of popular Internet services, such as on-demand video streaming, web browsing, and file downloading. The dataset, comprising over 220,000 measurements collected from two different vendor terminals and various mobility test modes, was obtained from a major French mobile operator in the Ile-de-France region over a six-month period in 2021. Various models from the literature are then implemented to estimate the QoE in terms of user Mean Opinion Score (MOS), utilizing features at both the radio and application levels. Furthermore, a detailed analysis of the collected dataset is conducted to identify the root causes of poor performance, revealing that radio provisioning issues are not the sole factor contributing to anomalies. Lastly, the chapter explores the prediction accuracy of HD video streaming services, considering factors such as launch time, bitrate, and MOS, based solely on physical indicators. The analysis encompasses both plain-text and encrypted traffic within different mobility modes, providing valuable insights into QoE optimization in mobile networks.