Personal comfort models based on a 6-month experiment using environmental parameters and data from wearables

Area of Science:

• Building Science
• Human-Computer Interaction
• Environmental Psychology

Background:

• Personal thermal comfort models are advancing building occupant perception prediction.
• Previous models faced limitations in data duration and environmental diversity.
• Wearable technology offers new avenues for real-time thermal comfort assessment.

Purpose of the Study:

• To outline a longitudinal field study on personal thermal comfort.
• To train and test machine learning models for predicting individual thermal preferences.
• To assess the impact of environmental and physiological variables on thermal perception.

Main Methods:

• A 180-day longitudinal field study with 20 participants.
• Collection of over 1080 Right-Here-Right-Now surveys per participant via smartwatch.
• Matching survey data with indoor environmental and physiological measurements (skin temperature, heart rate).

Main Results:

• Personal comfort models achieved a median prediction accuracy of 0.78 (F1-score).
• Skin, indoor, near-body temperatures, and heart rate were key predictive variables.
• Approximately 250-300 data points per participant were found necessary for accurate prediction, with strategies to reduce this number.

Conclusions:

• Wearable devices significantly benefit the prediction of thermal preferences.
• Quantitative evidence supports improving personal comfort model accuracy.
• Study findings validate previous research and highlight effective data collection strategies.