Abstract: The Metropolitan Area of Lima-Callao (MALC) stands out as one of the most environmentally challenged regions in Latin America and the Caribbean (LAC), characterized by elevated concentrations of the PM2.5 pollutant. This study aims to refine the accuracy of PM2.5 measurements obtained through a low-sensor instrument, employing machine learning (ML) technique with reference data provided by the National Service of Meteorology and Hydrology of Peru (SENAMHI) in the MALC. To train and evaluate the ML model, we utilized hourly PM2.5 concentrations from SENAMHI’s reference measurements and those from the low-sensor instrument operated by the Urban Transportation Authority for Lima-Callao (ATU) during the period from July 16th, 2023, to August 26th, 2023. The dataset was split into training (80%) and testing (20%) sets to construct a linear regression model using the scikit-learn package in Python. In the experimental results, the ML model demonstrated substantial improvement in Bias, reducing it from -12.32 to -4.11 when comparing PM2.5 concentrations measured by the low-sensor instrument against SENAMHI’s reference data. This enhancement was observed in the simulated PM2.5 concentrations generated by the ML model for the period spanning August 27th to August 31st, 2023. These findings underscore the potential of ML technique in refining air quality measurements and addressing pollution concerns in the MALC region.

Keywords: PM2.5 concentrations, Low-Sensor Instrument, Machine Learning (ML), Lima-Callao Metropolitan Area.