Data-driven model for predicting indoor air quality and thermal comfort levels in naturally ventilated educational buildings using easily accessible data for schools

Abstract

Modeling indoor air quality and thermal conditions in educational buildings is significant for protecting students' health, well-being, and productivity. The predictive models in existing studies were mainly built and applied in controlled environments with HVAC systems. These models did not involve occupant-related factors, had limited scope in a single building or space, and required indoor environmental monitoring data for the model input. This limits the applicability and generalization ability of the model in a large number of schools relying on natural ventilation, where indoor air quality and thermal conditions are significantly affected by occupants’ activities and ventilation practices. Hence, this paper proposes a methodology to develop a data-driven model for predicting the level of indoor air quality and thermal comfort in naturally ventilated educational buildings, and identifies the key influential factors. The model was developed using the class-weighted random forest algorithm with data collected from a large measurement campaign. The developed model demonstrated good accuracy, generalization ability, and robustness. The analysis concluded that occupancy, windows and doors operation, and outdoor environmental parameters are key factors must be involved, whereas building characteristics have no practical contribution to the prediction. The model inputs are easily accessible data for schools. Once the model is developed with data collected from an initial measurement campaign in representative educational buildings, it can be used in all local schools without requiring monitoring sensor networks, thereby rendering a “cost-effective” way of assessing indoor air quality and thermal comfort to help relevant stakeholders improve building management practices in schools.