Optimization of spatio-temporal ozone (O3) pollution modeling using an ensemble machine model learning with a swarm-based metaheuristic algorithm.

The future of ozone (O3) pollution presents significant environmental and public health challenges worldwide. High O3 levels can harm respiratory health, exacerbating conditions such as asthma and increasing the risk of cardiovascular diseases. Addressing these challenges requires advanced spatio-temporal modeling techniques to assess and predict O3 pollution levels accurately. This research fills a crucial gap in current modeling approaches by proposing a novel methodology that integrates an ensemble machine-learning algorithm with swarm-based metaheuristic optimization algorithm. The study uses surface-based ozone data and data for 14 environmental factors for Tehran, Iran between 2018 and 2022 to develop a spatio-temporal model of O3 pollution. The ensemble machine learning algorithm was selected as the base model, specifically the Random Forest (RF). To enhance its performance, a metaheuristic algorithm (Cuckoo search (CS) algorithm) was utilized for optimization. The evaluation of ozone risk maps, measured using the Receiver Operating Characteristic (ROC) curve, demonstrated strong performance across seasons. Specifically, the accuracy of the O3 risk map was 95.2 % for autumn, 97 % for spring, 96.7 % for summer, and 95.7 % for winter. This research provides actionable information for policymakers and public health officials to mitigate the impacts of O3 pollution on human health and the environment.