Pemodelan Deret Waktu ISPU Pm2.5 Menggunakan Metode Mlr dan Arima Di Kota Jakarta Pusat, Kota Malang dan Sidoarjo

Abstract

Air pollution from fine particulate matter PM2.5 poses a critical threat to urban environmental health. This study evaluates the predictive performance of Multiple Linear Regression (MLR) and Autoregressive Integrated Moving Average (ARIMA) models in forecasting the daily Air Pollutant Index (API) of PM2.5 across Central Jakarta, Malang, and Sidoarjo. Utilizing daily historical records from January 2024 to December 2025, the research analyzes the impacts of local meteorological drivers and projects a four-month future forecast (January–April 2026). Missing data points were handled by comparing MICE-Spline Interpolation against Listwise Deletion, while chronological data splitting (80:20, 70:30, 60:40) was applied to eliminate data leakage risks. The results demonstrate that the MLR model under a 60:40 splitting ratio achieved the highest predictive accuracy in all regions, yielding the lowest error metrics MAE, MAPE and positive Adjusted R-Squared values. Conversely, the ARIMA models suffered from severe structural overfitting during out-of-sample testing due to a mean-reversion limitation that failed to capture extreme pollution surges. The dominant meteorological factors varied by location: southwestern wind direction in Central Jakarta (t-statistic = 2.64), relative humidity in Malang’s mountainous basin (t-statistic = 3.26), and average temperature in coastal Sidoarjo (t-statistic = 2.11). The 2026 forecast projects stable, safe air quality in Malang ("Good" to "Moderate"), while Central Jakarta and Sidoarjo show dynamic risks of periodically reaching "Unhealthy" status.