Comparison of calibration models for low-cost PM2.5 sensors in high-concentration occupational environments

Low-cost PM2.5 sensors are increasingly used for ambient and indoor air quality monitoring, but their performance in high-concentration occupational environments remains uncertain. This study compares linear regression (LR), polynomial regression (PR), and random forest (RF) calibration models for low-cost PM2.5 sensors operating in a controlled high-concentration chamber. This study assessed the calibration performance of low-cost aerosol sensors (LCS) by employing LR, PR, and RF models across various aerosol (PM2.5) concentration ranges. Conventional LR models demonstrated solid performance when dealing with lower concentrations (0 to 150 µg/m3), but their accuracy diminished when confronted with higher concentrations. On the other hand, RF models continuously demonstrated better performance over all concentration ranges, making them potentially more appropriate for occupational situations with higher and fluctuating aerosol levels. PR models exhibited intermediate performance, surpassing that of LR but falling short of the robustness demonstrated by RF. The study highlights the importance of using modern calibration methods such as RF in locations with high concentrations to enable accurate aerosol monitoring using LCS. These results could lead to the development of efficient and cost-effective methods for monitoring air quality using LCS for high concentration environments.