Evaluating Indoor Low-Cost Particle Sensors: Algorithmic Insights and Calibration Approaches

Abstract

Purpose of Review

Low-cost particulate matter (PM) sensors are increasingly used for indoor air quality monitoring due to their affordability and ease of deployment. However, concerns persist regarding the reliability of their built-in processing functions and the accuracy of their data. This study evaluates the performance of 30 Plantower PMS5003 sensors across three distinct indoor environments—Ex_Normal (typical occupied office space), Ex_Incense (space with anthropogenic particle emissions), and Ex_Bushfire (space affected by outdoor air pollution). The primary aim is to improve data reliability by examining the sensors’ internal processing algorithms and identifying effective calibration models.

Recent Findings

Piecewise linear regression analysis revealed two key internal functions within the sensor: one for converting particle number to mass and another for adjusting based on particle type. Three calibration models—Log-Linear (LN), non-Log-Linear (nLN), and Random Forest (RF)—were evaluated. All models showed improvements over raw sensor data in terms of coefficient of determination (r²), root mean square error (RMSE), mean normalized bias (MNB), and coefficient of variation (CV), with particularly notable enhancements in RMSE (up to 64%), MNB (up to 70%), and CV (over 50%).

Summary

Although all three calibration models significantly improved data quality, no substantial differences were observed among them. The LN model is recommended for its simplicity and comparable performance. These findings contribute to improving algorithmic processing in low-cost sensors and offer practical guidance for end-users seeking to enhance sensor reliability in indoor air quality monitoring applications.