Evaluation of machine learning models to estimate particulate matter (2.5) in oil producing Arab countries: Kuwait as a case study

Abstract

Air pollution is one of the most serious environmental issues facing the State of Kuwait. The environment and the population's health depend on the state's ability to maintain air quality. Most air pollution is caused by the increasing population, increased human activities, and increased vehicle ownership. Reducing pollution and preserving the environment therefore depends on managing environmental pollutants. This research aimed to create a reliable method for assessing the concentrations of Particulate Matter 2.5 (PM_2.5; i.e., particles with a diameter less than 2.5 μm (μm)), one of the most harmful pollutants. Most of the existing studies on air quality prediction in Kuwait or similar areas face some challenges in the model accuracy, lack of datasets, and environmental variation. To address these challenges, four machine learning models: neural network models, support vector machine algorithms, generalized additive models, and random forests were used. The models were trained in four scenarios: daily or monthly data and data divided by city or aggregated. The results showed that the random forest model outperformed the others in predicting PM_2.5 concentration in all scenarios, with Coefficient of determination (R²) over 0.85, root mean squared error (RMSE) under 6.25 μg/m³ and mean absolute error (MAE) under 6.17 μg/m³. This study can be used as a guide for choosing the best model for estimating daily PM_2.5 concentrations in Kuwait and other places with the same climate.