Calibration of a low-cost PM2.5 monitor using a random forest model

Background: Particle air pollution has adverse health effects, and low-cost monitoring among a large population group is an effective method for performing environmental health studies. However, concern about the accuracy of low-cost monitors has affected their popularization in monitoring projects. Objective: To calibrate a low-cost particle monitor (HK-B3, Hike, China) through a controlled exposure experiment. Methods: Our study used a MicroPEM monitor (RTI, America) as a standard particle concentration measurement device to calibrate the Hike monitors. A machine learning model was established to calibrate the particle concentration obtained by the low-cost PM2.5 monitors, and ten-fold validation was used to test the model. In addition, we used a linear regression model to compare the results of the machine learning model. A calibration method was established for the low-cost monitors, and it can be used to apply the monitors in future air pollution monitoring projects. Results: The values of the random forest model calibration results and observations were more condensed around the regression line y = 0.99x + 0.05, and the R squared value (R-2 = 0.98) was higher than that for the linear regression (R-2 = 0.87). The random forest model showed better performance than the traditional linear regression model. Conclusions: Our study provided an effective calibration method to support the accuracy of low-cost monitors. The machine learning method based on the calibration model established in our study can increase the effectiveness of future air pollution and health studies.