<![CDATA[Fine particulate matter (PM2.5) is a key indicator of air quality, profoundly impacting human health and the environment. Pontianak City, located in the equatorial tropics of Indonesia, faces recurring air quality challenges driven by local meteorological variability and seasonal biomass burning in the surrounding regions. This study developed an hourly PM2.5 prediction model using an ensemble machine learning approach that integrates Random Forest (RF) and Extreme Gradient Boosting (XGBoost) algorithms. The dataset comprised hourly observations aggregated into hourly values from August 2024 to August 2025, including PM2.5 concentrations and three meteorological predictors: temperature, relative humidity, and atmospheric pressure. Correlation and mutual information analyses revealed that temporal features, notably lagged PM2.5 and 24-hour rolling averages, exerted the strongest influence on current PM2.5 levels, whereas meteorological variables contributed marginally in a nonlinear manner. Model evaluation demonstrated that the Hybrid Ensemble (Stacking RF–XGB) achieved the best performance with R² = 0.72, MAE = 9.71 µg/m³, and RMSE = 22.79 µg/m³, outperforming individual models (RF: R² = 0.69; XGBoost: R² = 0.65). The hybrid model effectively captured temporal fluctuations and extreme pollution events, offering improved robustness and generalization. These results highlight the potential of ensemble-based machine learning to enhance short-term air quality forecasting systems in tropical regions, providing valuable support for public health management and early warning strategies in Pontianak and similar urban environments in the future.]]>
