Análise não supervisionada para inferência de qualidade de experiência de usuários residenciais

Assessing the quality of experience of residential users is of great interest to ISPs. However, obtaining perceived QoE is costly, making it difficult to use supervised classifiers. This paper proposes a method based on unsupervised machine learning that detects statistical patterns in time series from the detection of change points and the spatiotemporal correlation of QoS measurement results. We exemplify the application of the method to a set of actual data, showing that the model results reflect a user QoE metric obtained from technical calls made to the call center. Finally, we evaluated the accuracy of the online execution of the method.