Evaluation of various machine learning prediction methods for particulate matter PM_10 in Kuwait

ir pollution poses a serious threat to public health and for the environment, thus predicting air quality is very crucial for the health and well-being of individuals and the environment. Economic development drives rapid industrialization and urbanization, which are significant sources of air pollution in developing countries. Kuwait’s rapid urbanization and vehicular traffic, along with dust storms, make it a prime location for research for environmental pollution. Keeping this in view, a study was designed to evaluates various machine learning prediction methods for particulate matter concentrations ( PM_10 ) in Kuwait. The prediction models were developed using three different algorithms, including k-nearest neighbor (KNN), artificial neural network (ANN) and support vector regression (SVR). The models were developed using a 3-year dataset collected by Kuwait Environmental Public Authority (K-EPA) for two stations selected in this study (Al-Ahmadi and Al-Salam). The performance of the models was evaluated using various metrics, including Mean Biased Error (MBE), Root Mean Squared Error (RMSE), Normalized Root Mean Squared Error (nRMSE) and Coefficient of Determination ( R^2 ). The results show that both stations experienced severe air quality issues and that particulate matter concentrations (PM10) are strongly influenced by the different meteorological and pollutant variables. The findings show that for the Al-Ahmadi location, artificial neural networks (ANN) ( R_cal^2 = 0.885, R_val^2 = 0.775) and K-nearest neighbor (KNN) ( R_cal^2 = 0.895, R_val^2 = 0.613) were good, while for Al-Salam, KNN ( R_cal^2 = 0.945, R_val^2 = 0.715) was a better choice to predict PM_10 . These models can be used by the decision-makers to impose pollution controls, evaluate policies, or plan targeted actions to reduce particle matter.