Towards accurate models for predicting smartphone applications' QoE with data from a living lab study

Area of Science:

• Human-Computer Interaction
• Software Engineering
• Mobile Computing

Background:

• Smartphones are essential tools, but mobile application performance often degrades user experience.
• Existing Quality of Experience (QoE) prediction models for mobile apps primarily use limited qualitative or quantitative data.

Purpose of the Study:

• To model and predict the Quality of Experience (QoE) for mobile applications on WiFi and cellular networks.
• To provide actionable recommendations for developers to create QoE-aware applications.
• To investigate the impact of user context and expectations on QoE prediction.

Main Methods:

• Conducted a 4-week in-the-wild study with 38 Android users, collecting 6086 qualitative and quantitative ratings.
• Utilized a smartphone logger (mQoL-Log) to gather background data including network info, user activity, and battery status.
• Applied data aggregation and feature selection to train predictive QoE models, incorporating user expectations and on-device features.

Main Results:

• Model performance improved significantly with ratings collected within 14.85 minutes of app usage.
• Including user expectations as a feature boosted predictive model performance.
• An on-device model using only smartphone features showed lower performance compared to full-feature models.
• Surprisingly, user physical activity emerged as a more critical predictor than network Quality of Service (QoS) or app name.

Conclusions:

• User expectations and contextual factors like physical activity are crucial for accurate mobile QoE prediction.
• Recommendations are provided for developers to design more QoE-aware mobile applications.
• The study highlights the importance of considering diverse data sources for robust QoE modeling.