Quantifying Indoor Air Quality Determinants in Green-Certified Buildings Using a Hybrid Machine Learning Method: A Case Study in Florida

Abstract

This study investigates the indoor air quality (IAQ) conditions in green-certified buildings and examines the factors influencing them. An integrated IoT sensing system was implemented indoors and outdoors to assess the levels of particulate matter, nitrogen dioxide, and ozone at five Leadership in Energy and Environmental Design (LEED)-certified and five non-LEED educational buildings in Central Florida. Building-related characteristics were collected through walk-through surveys, BACnet systems, and construction drawings. An algorithm model based on support vector machine (SVM) and nonnegative matrix factorization (NMF) was developed to analyze the features of pollutants and the relative contribution of different influencing factors. The findings reveal that concentrations of target pollutants are generally lower in LEED buildings compared to non-LEED buildings. Although IAQ influencing factors are generally similar between LEED and non-LEED buildings, the weighted contribution ratios of specific factors, particularly for indoor nitrogen dioxide and ozone, vary significantly. The concentration of pollutants in non-LEED buildings is more susceptible to adverse environmental factors. The SVM-NMF model demonstrates significant advantages in nonlinear feature extraction and handling multicollinearity issues. It surpasses multiple linear regression and backpropagation neural network models in analyzing multidimensional indoor air data by 26.9% and 18% (p < 0.001), respectively. The robustness of the model was validated through fit comparison, cross-validation, and residual analysis. This study provides a foundational information base and effective technical means for subsequent research on IAQ management.