Smart Urban Pollution Management Using Deep Learning Techniques for NO2 Prediction

Abstract

[Abstract] The rapid pace of urbanization and the expansion of economic activities have significantly contributed to rising levels of air pollution, posing serious threats to human health and the environment. Among the key pollutants, Nitrogen Dioxide (NO2) stands out due to its strong association with respiratory and cardiovascular diseases and its role in exacerbating climate change. NO2 is predominantly emitted through fossil fuel-powered transportation and industrial processes. At the economic level, air pollution entails high costs, such as lost productivity in key sectors and increased public spending on environmental remediation initiatives. In this context, Artificial Intelligence (AI) tools, such as Machine Learning (ML) and Deep Learning (DL), are now indispensable for analyzing environmental data and offering accurate predictions about pollution levels. Hence, this work focuses on the design, tuning, and evaluation of five DL models, namely Multi-Layer Perceptron (MLP), LSTNet, Temporal Convolutional Networks (TCN), Transformers, and Transformers combined with Long Short-Term Memory (LSTM), to forecast NO2 concentrations in the city of Porto up to two days in advance. The results reveal that the MLP model demonstrated the highest performance, achieving a Root Mean Square Error (RMSE) of 7.65 µg/m3, outperforming more complex architectures. These findings underscore the effectiveness of DL in pollutant forecasting, contributing to more informed decision-making and air pollution mitigation strategies.