Prediction of PM10 Concentration in South Korea Using Gradient Tree Boosting Models

Abstract

Particulate matter (PM) is a term generally used for very small particles and liquid droplets in the atmosphere. PM10 is the particle pollution with diameter less than or equal to 10 micrometers. Exposure to particle pollution is a public health hazard which leads to serious diseases such as asthma, bronchitis and even cancer; especially in elderly, children and sensitive people. It is crucial to predict the concentration of PM before-hand so that people can take precautionary measures and avoid the hazardous impact of pollution. These days the gradient boosting is one of popular methods in regression and classification tasks. In this study, we predict the PM10 concentration using Extreme Gradient Boosting (XGBoost) and Light Gradient Boosting Machine (LightGBM) algorithms after combining meteorological, emission rate data and output features of Community Multi-Scale Air Quality (CMAQ) model. All the missing values are removed because handling them is quite challenging and requires feature engineering. The results show that XGBoost performs better than LightGBM in terms of prediction estimation with the RMSE of 12.846; but takes longer to train and tune the model's parameters. RMSE of LightGBM is 12.9066, which is slightly higher; but on the contrary, it is 29 times faster than XGBoost.