Automatic calibration methodology using machine learning with low-cost and meteorological sensors

In the last years several low-cost sensors (LCS) were installed in Lima, however, the accuracy, reliability, comparability and standardization compared with robust Senamhi sensors is a problem. The calibration helps to align the readings of the LCS with the known reference concentration sensors and values, on the other hand the reliability helps to reduce the variations in readings caused by factors like temperature changes, humidity and sensor degradation; besides, the comparability allows to compare the data across different monitoring systems with the standardization making it easier to interpret and compare data registered from different LCS. Therefore, an evaluation of LCS and Senamhi sensors with several algorithms were tested with five algorithms; for the calibration of the PM2.5 with EPA standards set limits for the concentration of fine particles with the calibration of the sensors with Random Forest, in the air with the highest value of correlation with 91.92%, with MAE 2.462 µ g/m
³
, RAE of 35.7328%, and RRSE of 49.0361%.