Decoding Living Lab sensing system through Bayesian networks: The preferable working space targeting comfort and productivity

Workplace environmental conditions significantly influence workers' well-being, health, and productivity. The understanding of the interaction between environmental factors across multiple domains can improve occupants' satisfaction and indoor conditions. This study aims to identify optimal office environmental conditions by analysing comfort demands and multi-domain interplay in a dynamic context. A year-long experimental campaign was performed in a Living Lab comprising five offices, monitoring indoor and outdoor environmental parameters and gathering occupants' feedback through surveys. Gaussian Naïve Bayes technique was applied to develop probabilistic models that identified optimal conditions for comfort and satisfaction, including self-perceived productivity as combined effect analysis. Participants showed more acceptability of warmer conditions in cooling seasons. Probabilistic models aligned with the current standards, indicating optimal temperatures of 22°C-24 °C for heating seasons and 23°C-24 °C for cooling seasons. Visual comfort was affected by the balance of natural and artificial light, with higher visual discomfort when the former was limited during cooling seasons. Temperatures higher than 22 °C demonstrated increased “stale air” discomfort, potentially linked to respiration and sweating, even without elevated CO₂ levels. Self-perceived productivity decreased in temperatures higher than 24 °C and CO2 concentrations exceeding 800 ppm. Other factors and more comprehensive measurements, together with monitoring of physiological signals should be included in future studies, allowing the creation of guidelines for more comfortable office places. These findings offer valuable insights for enhancing workplace human-centric standards and regulations globally. They have the potential to shape policies that foster more sustainable productive environments for workers’ wellbeing worldwide.