Section: New Results

Predicting the effect of home Wi-Fi quality on Web QoE

Participants: Diego Neves da Hora, Renata Teixeira, Karel Van Doorselaer (Technicolor), Koen Van Oost (Technicolor)

We developed a model that predicts the effect of Wi-Fi quality on Web QoE, using solely Wi-Fi metrics commonly available in commercial APs. We trained our predictor during controlled experiments on a Wi-Fi testbed and assess its accuracy through cross-validation, obtaining an RMSE of 0.6432 MO, and by applying it on a separate validation dataset, obtained on an uncontrolled environment, finding an RMSE of 0.9283. Finally, we apply our predictor on Wi-Fi metrics collected in the wild from 4,880 APs over a period of 40 days. We find that Wi-Fi quality is mostly good for Web—in more than 60% of samples Wi-Fi quality does not degrade Web QoE. When we consider average complexity Web pages, however, Wi-Fi quality degrades Web QoE in $11\%$ of samples. Moreover, we saw that $21\%$ of devices present more than $20\%$ of poor Web QoE samples, with $5\%$ of these showing highly intermittent QoE degradations, which are particularly hard to diagnose, indicating the need for a long-term monitoring approach to detect and fix problems.