Air pollution in urban areas like Surabaya poses signifcant risks to public health. Accurate forecasting ofAir Pollution Index (ISPU) parameters is essential for early warning systems. Previous studies often overlook temporaldependencies by focusing on instantaneous correlations. This study proposes a temporal forecasting framework usingExtreme Gradient Boosting (XGBoost) with autoregressive features and Long Short-Term Memory (LSTM) networks. Datafrom 2021 to 2024 were processed using a sliding window approach (lags) to capture historical patterns. Results indicatethat while XGBoost provides robust baseline predictions, LSTM’s ability to retain long-term dependencies yields superiorstability in capturing complex pollutant fluctuations. Although RMSE and MAE values were signifcantly improved throughtemporal modeling, the moderate R2 scores suggest that external meteorological factors, such as wind speed and humidity,remain critical latent variables. This study contributes a methodological benchmark for urban air quality monitoring usinghybrid machine learning approaches.
Herlambang et al. (Fri,) studied this question.