Contribution of influential factors on PM2.5 concentrations in classrooms of a primary school in North China: A machine discovery approach

Abstract

Indoor air quality is crucial to the physical and psychological health of students because they spend a large portion of their time in classrooms. Indoor PM2.5 concentrations are affected by various complex factors, such as outdoor PM2.5, temperature, and relative humidity. However, how and to what extent these factors contribute to PM2.5 concentrations remains unclear. Recently, black-box (e.g., neural networks) and white-box (e.g., linear regression) models have been applied to analyze their contributions. However, these models either have low interpretability or are insufficient for revealing complex nonlinear relationships. To address these issues, indoor PM2.5 concentrations in ten classrooms were monitored for twelve months at a primary school in North China, and the contribution of various influential factors was calculated using partial derivatives based on this approach. This method can automatically identify linear and nonlinear relationship models from a large amount of data without a prior hypothesis. Experimental results showed that: (1) The contribution of each influential factor to indoor PM2.5 varied greatly in the 12 months based on the identified relationships. In addition, no single relationship pattern was identified that could universally describe the factors influencing indoor PM2.5. (2) The method developed in this study can automatically identify linear and nonlinear relationship models with white-box structures from a large amount of data without a prior hypothesis, which is more interpretable and applicable than other baselines. (3) Targeted solutions can be drawn by the contribution of each factor and three types of relationship curves (i.e., N-shaped, M−shaped, and W-shaped), the latter of which is the Pareto front revealed by our results. The study offers informative results regarding the temporal dynamics of indoor PM2.5 concentrations and quantitative contributions of influential factors based on timely long-term data, which promotes the understanding of changes in indoor PM2.5 concentrations. This also demonstrates the promising potential of using low-cost online sensors to conduct informative measurements that guide schools to more effectively improve classroom air quality.