Improved \(NO_2\) Prediction Using Machine Learning Algorithms

Improved air pollution management approaches are required to ensure better air quality and tackle climate change. The ability to accurately forecast air quality, particularly the concentration of \(NO_2\) in the air, is crucial especially for urban settings due to direct health implications. Various machine and deep learning models have been used for air quality prediction. However, the application of these approaches on \(NO_2\) concentration levels, focusing on specific cities and specific climatic conditions, has been investigated on a limited scale. In this study, the performance of commonly used algorithms, such as Random Forest Regressor, Light Gradient-Boosting Machine (LightGBM), Extreme Gradient Boosting (XGBoost), and Long Short-Term Memory (LSTM) networks for predicting \(NO_2\) concentration levels based on time series and meteorological data including climatic conditions is assessed. Further, ensemble modeling techniques are evaluated as a voting regressor using Random Forest, LightGBM, and XGBoost as base models for improving the prediction of \(NO_2\) concentration levels. Each model was evaluated using cross-validated (5-fold) Mean Absolution Error and Root Mean Square Error metrics with LSTM emerging as the best-performing model.