Accurate predictions of air pollutant PM10 concentrations are essential for crafting effective air quality management strategies. This study compares three decision tree ensemble models—Random Forest (RF), Extra Trees, and Extreme Gradient Boosting (XGBoost)—to forecast daily PM10 levels in Thiruvananthapuram, India. By integrating meteorological data and air pollutant variables, this study aims to enhance both the accuracy and interpretability of urban air pollution dynamics. Spearman correlation analysis is employed to analyse the relationships between PM10 and the various input features. The predictive performance of the ensemble models is evaluated using Root Mean Squared Error (RMSE) and Coefficient of Determination (R²). The Extra Trees model demonstrates superior predictive performance, achieving an R² of 0.945 and an RMSE of 8.174 μg/m³. The model-agnostic interpretability method SHapley Additive exPlanations (SHAP) demonstrates that PM2.5, NH3, NO2, and O3 have a major impact on PM10 forecasts. Additionally, it reveals that meteorological conditions, particularly rainfall and relative humidity, play a crucial role in determining PM10 concentrations. This research highlights the potential of machine learning techniques, especially when combining the Extra Trees model with SHAP, to assist local governments in strategic planning and air quality management efforts. Although temporal coverage limits are acknowledged, this study offers useful information to environmental agencies and policymakers looking for data-driven strategies to reduce air pollution.