Comprehensive evaluation of machine learning models for real-world air quality prediction and health risk assessment by AirQ+
Ebru Koçak
Abstract
Abstract This study extensively examines five distinct machine learning models used to predict hourly air particulate matter concentrations. The study used real-world data, including pollutant levels and various meteorological parameters, for model training and evaluation, making the study more reliable and effective. The study focused on capturing short-term trends in pollutant concentrations and meteorological conditions. Results showed varied model performances. The Ridge Regression model exhibited a moderate R 2 value of 0.44 for PM 2.5 prediction and an impressive R 2 of 0.91 for PM 10 prediction. Support Vector Regression showed strength in PM 2.5 prediction (R 2 = 0.83) but faced challenges in forecasting PM 10 . Random Forest and Extra Trees Regression demonstrated robust overall performance, particularly in PM 10 forecasting (R 2 = 0.75). Extreme Gradient Boosting displayed competitive results for both PM 2.5 and PM 10 (R 2 = 0.80 and 0.81). Each model's identified strengths and limitations provide valuable insights for air quality management, offering a foundation for future research and the development of machine learning models in the continuous pursuit of accurate and timely air quality predictions. The AirQ+ model was used to estimate the health effects of PM 2.5 exposure and predict the long-term mortality rates associated with PM 2.5 . The average estimated attributable proportion for all years is 10.2% (with a range of 6.5% to 13.2%). The results show differing trends in estimated mortality rates, underscoring the need for targeted interventions to reduce the public health risks associated with exposure to polluted air.