Comparative Evaluation of Low-Cost CO2 Sensors for Indoor Air Pollution Monitoring

In this paper, four low-cost CO2 sensors are evalu-ated for IoT-based indoor air pollution monitoring. Specifically, CO2 sensors SCD30, Prana Air, MHZ14, and T6713 are eval-uated against a standard reference Aeroqual S-500 device. The experiment was carried out in an indoor environment inside one of the labs in IIIT Hyderabad, India. It is shown that calibration is needed for some of these low-cost devices locally even though the sensors may be factory calibrated. For calibration, simple and widely-used machine learning algorithms are employed such as linear regression, least absolute deviation, random forest, support vector regression, and Gaussian regression. The parameters considered to assess the performance of these sensors are coefficient of determination
$(R^{2})$
, coefficient of variability
$(C_{v})$
, and root mean square error (RMSE). After calibration with a reference sensor, it is observed that these low-cost sensors operate well.