Mobile QoE Prediction in the Field

In Pervasive and Mobile Computing

Quality of experience (QoE) models quantify the relationship between user experience and network quality of service. With the exception of a few studies, most research on QoE has been conducted in laboratory conditions. Therefore, in order to validate and develop QoE models for the wild, researchers should carry out large scale field studies. This paper contributes data and observations from such a large-scale field study on mobile devices carried out in Finland with 292 users and 64,036 experience ratings. 74% of the ratings are associated with Wifi or LTE networks. We report descriptive statistics and classification results predicting normal vs. bad QoE in in-the-wild measurements. Our results illustrate a 20% improvement over baselines for standard classification metrics (G-Mean). Furthermore, both network features (such as delay) and non-network features (such as device memory) show importance in the models. The models' performance suggests that mobile QoE prediction remains a difficult problem in field conditions. Our results help inform future modeling efforts and provide a baseline for such real-world mobile QoE prediction.

The QoE App Rating Dataset

The dataset contains mobile (including LTE, Wifi, and HSPA+) app-level QoE data collected from a large (292 users with 64,036 experience ratings), diverse group of participants during Summer 2017. The participants installed a custom measurement app that both monitors network quality and prompts participants to rate their experiences with other apps during or after usage of those other apps. Refer to the publication for more information.

The dataset is free to use for non-commercial purposes with attribution under CC BY-NC 4.0.

QoE App Rating Dataset (CSV format, 4.2 MB zipped, 12.6 MB unzipped).


Not yet available


For questions and further information, please contact:

Benjamin Finley

benjamin.finley (at)

This work was supported by the EMERGENT Project.