Air Quality Prediction Using Ensemble Machine Learning Models with Environmental and Meteorological Features

Air pollution, especially fine particulate matter (PM2.5), poses serious health and environmental threats globally. Traditional models for air quality prediction often fall short in capturing the complex, dynamic nature of pollution due to their linear assumptions and lack of contextual information such as meteorological or human-activity patterns. This research work proposes an ensemble machine learning framework that integrates both environmental pollutant data and temporal features derived from timestamps to enhance the accuracy, robustness, and interpretability of PM2.5 prediction. The dataset used includes hourly air quality measurements from Delhi, consisting of pollutant concentrations (CO, NO, NO₂, O₃, SO₂, PM₁₀, NH₃) and timestamps. Data preprocessing involves parsing timestamps into structured datetime features (hour, day, month, weekday), handling missing values using mean-based imputation, and applying standard normalization. Two ensemble learning models—Random Forest Regressor (RF) and Gradient Boosting Regressor (GBR)—are trained to predict PM2.5 concentrations. Model performance is evaluated using MAE, MSE, RMSE, and R². Additionally, feature importance analysis from the Gradient Boosting model is conducted to enhance explainability. The Random Forest model achieved superior predictive performance, with an MAE of 9.386, RMSE of 15.265, and an R² score of 0.995, outperforming Gradient Boosting which yielded an MAE of 11.794, RMSE of 17.585, and R² of 0.994. These results significantly improve upon earlier baseline performances (e.g., MAE = 22.4), highlighting the impact of integrating temporal features and comprehensive preprocessing. Feature importance analysis further identified key contributors to PM2.5 concentration, improving interpretability. The proposed framework demonstrates that combining environmental and temporal features with ensemble models significantly enhances air quality prediction. The Random Forest model, in particular, proves effective in modeling complex, non-linear pollution behavior. The addition of feature explainability supports data-driven decision-making for environmental policy and real-time public health interventions.