ContextLabeler dataset: Physical and virtual sensors data collected from smartphone usage in-the-wild

Area of Science:

• Computer Science
• Human-Computer Interaction
• Ubiquitous Computing

Background:

• Understanding daily life activities through sensor data is crucial for context-aware systems.
• Existing datasets may lack the 'in-the-wild' authenticity needed for robust mobile solutions.

Purpose of the Study:

• To present a novel dataset capturing diverse daily life activities using smartphone sensors.
• To provide a valuable resource for developing and evaluating context-aware algorithms and protocols.

Main Methods:

• Conducted an 'in-the-wild' data collection campaign with 3 volunteers over two weeks.
• Utilized smartphone sensors to gather data on motion, running applications, nearby devices, and weather.
• Collected over 45,000 data samples, each with 1332 features and ground truth activity labels.

Main Results:

• A comprehensive dataset of smartphone sensor readings linked to user activities (e.g., working, dining, exercising).
• The dataset reflects natural user behavior without imposed constraints, ensuring ecological validity.

Conclusions:

• The 'in-the-wild' dataset is a significant resource for advancing context-aware computing in mobile environments.
• This data enables the creation and validation of adaptive systems that respond to real-life user situations.