Multi-sensor dataset of human activities in a smart home environment

Area of Science:

• Computer Science
• Artificial Intelligence
• Gerontology

Background:

• Smart home sensor data is crucial for developing intelligent systems to monitor occupants' activities, supporting independent aging in place.
• The increasing need for home monitoring solutions highlights the importance of accessible, real-world datasets.

Purpose of the Study:

• To describe a novel, unlabelled dataset of environmental sensor measurements from a smart home.
• To facilitate research in activity recognition and habit pattern analysis using data-driven and AI algorithms.

Main Methods:

• Collected millions of raw sensor data samples continuously at 1 Hz over six months (Feb-Aug 2020).
• Utilized various sensors: passive infrared, force sensing resistors, reed switches, photocell light sensors, temperature/humidity, and smart plugs.
• Focused on capturing user interactions with the environment, including movement, pressure, and appliance usage.

Main Results:

• The dataset comprises unlabelled time-series measurements from multiple environmental sensors.
• It captures detailed human activities of daily living through sensor interactions.
• Data spans six months, offering a rich resource for longitudinal habit analysis.

Conclusions:

• The unlabelled dataset is valuable for developing and testing data-driven and AI algorithms for activity and habit recognition.
• It allows investigation into algorithm performance on unlabelled data and extraction of long-term user behavior patterns.
• Enables detection of habit changes, potentially leading to timely interventions for individuals, such as those with dementia.