Residual-Based Air Quality Monitoring: A Hybrid XGBoost-AEWMA Control Chart for Detecting PM_2.5 Anomalies

The concentration of ambient particulate matter with an aerodynamic diameter less than or equal to 2.5 micrometers (PM_2.5) represents a critical environmental health indicator due to its capacity to penetrate deeply into the alveolar regions of the respiratory system and subsequently translocate into the circulatory system, increasing the risk of adverse health outcomes. Given the inherent complexity of PM_2.5 time series data - often characterized by nonlinear dynamics and autocorrelation - accurate and robust monitoring methodologies are essential. This study introduces a hybrid approach to PM_2.5 monitoring by combining the nonlinear predictive capabilities of the Extreme Gradient Boosting (XGBoost) algorithm with the sensitivity of the Adaptive Exponentially Weighted Moving Average (AEWMA) control chart applied to forecast residuals. Using daily PM_2.5 data from Jakarta (Jan 2024-Apr 2025), the model was trained and evaluated for prediction and anomaly detection. XGBoost effectively captured the nonlinear characteristics of the data, achieving a very low Mean Absolute Percentage Error (MAPE). Furthermore, the AEWMA chart exhibited strong sensitivity in detecting statistically significant deviations from predicted values, identifying potential air pollution events. Compared to an ARIMA model, XGBoost showed superior predictive performance. The comparative analysis across various forecasting and control chart combinations confirmed that the XGBoost-AEWMA framework significantly outperforms other configurations. This hybrid approach enhances anomaly detection in complex, autocorrelated datasets and provides a practical framework for early air quality warning systems and policy support.