An unsupervised machine learning approach for indoor air pollution analysis

Abstract

Exposure to indoor air pollutants is one of the most significant environmental and health risks people face, especially since they spend most of their time indoors. Therefore, evaluating indoor air pollution levels and comfort parameters is essential for achieving sustainable indoor air quality (IAQ). The main objective of this study was to identify patterns of indoor air pollution in two buildings with different characteristics located on a university campus in northeastern Mexico. We measured the concentration of particulate matter in fractions of 1.0 μm (PM₁), 2.5 μm (PM_2.5), and 10 μm (PM₁₀), as well as carbon dioxide (CO₂), carbon monoxide (CO), and ozone (O₃), along with the temperature and relative humidity in each microenvironment during the working hours of spring, summer, and autumn. Next, unsupervised machine learning was employed to identify behavioral patterns of air pollutants within the microenvironments. The K-means clustering algorithm was used to identify homogeneous microenvironments within the study area. We performed three clustering analyses per building: (1) considering all the variables in the dataset, (2)selecting the significant variables through principal component analysis (PCA), and (3) examining two time ranges within the working day. The robustness of the proposed approach was evaluated through a comparative analysis of the K-means, DBScan, and hierarchical algorithms, assessing their performance using the Davies–Bouldin index and Silhouette score metrics. Furthermore, the stability of the clusters over time intervals was assessed using the adjusted Rand index. Cluster analysis enabled us to identify microenvironments with maximum similarity and those that change groups, as their behavior depends on the time range. Consequently, grouping microenvironments into homogeneous IAQ classes is effective in accurately identifying spaces based on patterns related to their contamination levels and guiding actions to reduce pollution levels by zone or building.