Real-Time Source Apportionment of Particulate Matter from Low-Cost Particle Sensors Using Machine Learning

Low-cost sensors (LCS) have gained significant attention in recent years due to their application in urban air quality mapping, community monitoring networks, indoor air quality monitoring, personal exposure monitoring, and citizen science initiatives. This study has developed an integrated approach combining measurements from LCS and existing source apportionment (SA) results with machine learning (ML) algorithms to achieve real-time SA. Source contributions apportioned by Chemical Mass Balance (CMB) model and PM_2.5 as well as particle number concentration (PNC) in size bins (0–0.3 μm, 0.3–0.5 μm, 0.5–1 μm, and 1–2.5 μm) from LCS are acquired from May 2019 to February 2020 at Major Dhyan Chand National Stadium (NS), Delhi. The PNC in size bins was converted to mass (PM_{0 − 0.3}, PM_{0.3 − 0.5}, PM_{0.5 − 1}, PM_{1 − 2.5}) for respective sizes. The objective function is {S₁, S₂, S₃, …. S₈} = f {PM_{0 − 0.3}, PM_{0.3 − 0.5}, PM_{0.5 − 1}, PM_{1 − 2.5}, PM_2.5} where S₁, S₂, S₃, …. S₈ are the sources. Four ML algorithms, namely support vector regression (SVR), k-nearest neighbour (kNN), random forest (RF) and gradient boosting (GB), are applied for SA. GB performs the best among all algorithms with a train and test score (R²) of 0.82 and 0.75. The R² (in parentheses) between actual and predicted PM_2.5 for sources of biomass burning (0.92), dust (0.83), gasoline vehicle (0.75), diesel vehicle (0.78), coal combustion (0.70), waste burning (0.76), industrial (0.77) and secondary aerosol (0.89) indicate the acceptable performance of the model. The statistical t-test comparing the PM_2.5 contributions obtained from CMB and ML for each source indicates no significant difference (p > 0.05) except for dust and waste burning. This study demonstrated the ability of LCS to perform real-time SA with the help of an existing dataset. This cost-effective approach will provide rough estimations of the sources to regulatory agencies and policymakers for immediate action.