Particulate matter is known to have adverse effects on health, making it crucial to accurately gauge its concentration levels. While the recent advent of low-cost air sensors has enabled real-time measurement of particulate matter, discrepancies in concentrations can arise depending on the sensor used, the measuring environment, and the manufacturer. In light of this, we aimed to propose a method to calibrate measurements between low-cost air sensor devices. In our study, we introduced decision tree techniques, commonly used in machine learning for classification and regression problems, to categorize particulate matter concentration intervals. For each interval, both univariate and multivariate multiple linear regression analyses were conducted to derive calibration equations. The concentrations of PM10 and PM2.5 measured indoors and outdoors with two types of LCS equipment and the GRIMM 11-A device were compared and analyzed, confirming the necessity for distinguishing between indoor and outdoor spaces and categorizing concentration intervals. Furthermore, the decision tree calibration method showed greater accuracy than traditional methods. On the other hand, during univariate regression analysis, the proportion exceeding a PM2.5/PM10 ratio of 1 was significantly high. However, using multivariate regression analysis, the exceedance rate decreased to 79.1% for IAQ-C7 and 89.3% for PMM-130, demonstrating that calibration through multivariate regression analysis considering both PM10 and PM2.5 is more effective. The results of this study are expected to contribute to the accurate calibration of particulate matter measurements and have showcased the potential for scientifically and rationally calibrating data using machine learning.

Concentrations of fine particulate matter (PM2.5) and several of its particle constituents measured outside homes in Houston, Texas, and Los Angeles, California, were characterized using multiple regression analysis with proximity to point and mobile sources and meteorological factors as the independent variables. PM2.5 mass and the concentrations of organic carbon (OC), elemental carbon (EC), benzo-[a]-pyrene (BaP), perylene (Per), benzo-[g,h,i]-perylene (BghiP), and coronene (Cor) were examined. Negative associations of wind speed with concentrations demonstrated the effect of dilution by high wind speed. Atmospheric stability increase was associated with concentration increase. Petrochemical source proximity was included in the EC model in Houston. Area source proximity was not selected for any of the PM2.5 constituents' regression models. When the median values of the meteorological factors were used and the proximity to sources varied, the air concentrations calculated using the models for the eleven PM2.5 constituents outside the homes closest to influential highways were 1.5-15.8 fold higher than those outside homes furthest from the highway emission sources. When the median distance to the sources was used in the models, the concentrations of the PM2.5 constituents varied 2 to 82 fold, as the meteorological conditions varied over the observed range. We found different relationships between the two urban areas, illustrating the unique nature of urban sources and suggesting that localized sources need to be evaluated carefully to understand their potential contributions to PM2.5 mass and its particle constituents concentrations near residences, which influence baseline indoor air concentrations and personal exposures. The results of this study could assist in the appropriate design of monitoring networks for community-level sampling and help improve the accuracy of exposure models linking emission sources with estimated pollutant concentrations at the residential level.