Artificial Intelligence–Driven Solutions in Environmental Engineering: Advancing Sustainable Urban Air Quality Through PM2.5 Prediction Using Machine Learning

Abstract

Air pollution prediction is an integral component of environmental engineering, and effective management of air quality demands timely, accurate and explainable air quality forecasting to manage urban pollution. The concept of this research was to use artificial intelligence to predict the PM2.5 level using the required predictors PM10 and NO₂ as well as other pollutant, meteorological, temporal, wind-direction and site-specific variables from the Beijing Multi-Site Air Quality dataset. A quantitative supervised regression approach was followed, covering data preprocessing techniques such as data cleaning, handling of missing values, temporal feature engineering, encoding of categorical variables, selection of models, hyperparameter tuning, cross-validation, residual analysis, error analysis for various categories of pollution, feature importance analysis, and explainability using SHAP. A common data mining pipeline was used to evaluate the performance of five models (Linear Regression, Decision Tree, Random Forest, Gradient Boosting, XGBoost). Random Forest Regressor achieved maximum predictive performance in terms of high accuracy, low validation error, low residual bias and acceptable performance in all the pollution categories. The importance of PM10 was identified by the explainability analysis, while the other environmental factors also provided valuable information. The results indicate that explainable ensemble learning can provide a solid basis for AI-based PM2.5 prediction and assist in air-quality early warning, environmental monitoring, pollution-control and sustainable urban management.