Assessing the perception of overall indoor environmental quality: Model validation and interpretation

Abstract

The indoor environmental quality (IEQ) was determined by assessment of its various domains, among which the four principal ones are thermal, acoustic, visual and indoor air quality (IAQ). While many IEQ models have been established to develop scheme for assessing the overall IEQ, the lack of model validation significantly limited their applicability. This study aimed to provide new insights into how occupants evaluate the overall IEQ performance by establishing regression models and machine learning models that predict occupants’ overall satisfaction using their satisfactions with the four principal domains of IEQ. Accuracy and applicability of proposed models were examined on two different datasets. Prediction error increased 21% when models were generalized to new dataset collected from different buildings and populations, and error of machine learning models increased more significantly than regression models. The model performance varied across building types. Models trained on samples collected from office settings had better generalization to the similar environment, while increased error was identified for schools and residences. The acoustic environment had a greater impact on females than males, while the IAQ had a greater impact on males. The Shapley Additive Explanations was applied to interpret model predictions and measure contribution of the variables. The results revealed that the unsatisfying IEQ factor often yielded a dominant negative impact on the overall IEQ satisfaction, which can hardly be compensated by a higher satisfaction with another factor. Efficient strategy can be developed to improve overall satisfaction with IEQ based on findings of the present study.