Conveners
Leveraging Traffic Information for Air Pollutant Estimation: Leveraging Traffic Information for Air Pollutant Estimation
- Valeria Karina Legaria Santiago (Queen Mary University of London)
Description
Vehicular traffic is a major source of air pollution; however, the contribution of traffic information to machine-learning models for air pollutant estimation remains insufficiently characterised. To address this, we analysed two case studies: one measuring air pollution at ground level and using traffic information from nearby roads, and the other measuring air pollution at rooftop level, with broader spatial representativeness, and considering traffic information from roads within this area. We used ablation experiments to systematically assess the contribution of traffic information to estimation models by comparing their performance with and without traffic data across different monitoring sites, air pollutants, and levels of additional information from neighbouring air-quality monitoring stations. During this seminar, we will discuss how tree-based ensemble models were trained to estimate air pollutant concentrations using nested cross-validation with time-series splits to account for the temporal structure of the data and Optuna for hyperparameter optimisation, and how SHAP (SHapley Additive exPlanations) was used to quantify the contribution of the variables of interest to the models' predictions.