The effect of preprocessing data on the forecast of PM2.5

In recent years, Jakarta’s air quality has deteriorated significantly, with PM2.5, far exceeding the WHO standards. This study aims to develop an accurate PM2.5 prediction model to support air quality management. Using hourly PM2.5 data from January 2021 to January 2024, we employed the Double Seasonal Holt-Winters method, capturing daily and weekly patterns. Four imputation scenarios are proposed to address the missing data. Forecasting was performed using Holt-Winters (HW) and Double Seasonal Holt-Winters (DSHW) methods, evaluated by Root Mean Square Error (RMSE) for 24 and 168 future observations. The results showed that the HW model performed better for short-term forecasts, whereas the DSHW model was more effective for long-term predictions. These findings offer valuable insights for PM2.5 emission reduction strategies and underscore the need for appropriate forecasting models based on forecast duration.