Forecasting Air Quality: A Comparative Study of LSTM, Prophet, and XGBoost

Accurate multi-step air quality forecasting is crucial for public health management in urban environments like DKI Jakarta. This study presents a comparative analysis of three prominent forecasting models: Long Short-Term Memory (LSTM), Prophet, and Extreme Gradient Boosting (XGBoost). Utilizing daily ISPU data from 2021 to 2023 for six key pollutants (PM10, PM2.5, SO2, CO, O3, NO2), the models were evaluated for their 60-day forecasting performance using Mean Absolute Error (MAE) and Root Mean Squared Error (RMSE), alongside computational efficiency metrics. The results indicate that the LSTM model, implemented using a direct multi-step strategy, achieved the highest predictive accuracy for the majority of pollutants, while XGBoost performed best for CO. This study addresses a critical gap by providing a direct, standardized multi-step evaluation of these distinct models on a localized, tropical urban dataset, offering a clear benchmark for future implementations. Considering the practical trade-off between accuracy and computational cost, Prophet emerges as a highly viable and efficient alternative, balancing reasonable predictive capabilities with significant advantages in speed and resource utilization.