Development of a machine learning model for air quality estimation considering key performance indicators for an Indian urban region

Abstract

This paper aims to develop a Machine Learning (ML)-based framework for estimating urban air quality using Key Performance Indicators (KPIs) influencing pollution levels in Delhi, India. The considered factors include geographic coordinates, population density, time-based traffic density, land-use type (residential, commercial, industrial), meteorological conditions, and seasonal variation. Ten ML and Deep Learning (DL) models were evaluated: Extreme Gradient Boosting (XGBoost), Categorical Boosting (CatBoost), Light Gradient Boosting Machine (LightGBM), Random Forest (RF), Decision Tree (DT), Multivariate Adaptive Regression Splines (MARS), FT-Transformer, Artificial Neural Network (ANN), TabNet, and CNN-LSTM. Model performance was assessed using R2, Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and Symmetric Mean Absolute Percentage Error (SMAPE) metrics, along with deployment feasibility for practical implementation. Among all models, XGBoost achieved the best performance with R2 = 0.8309, RMSE = 54.2866, MAE = 37.4963, and SMAPE = 20.9743, followed by LightGBM, CatBoost, and RF. Deployment analysis identified CatBoost and TabNet as the most suitable models for edge deployment owing to their good deployment feasibility, while MARS exhibited the lowest predictive accuracy among the evaluated AI models.