Pengembangan Model Prediksi Risiko Kesehatan Akibat Paparan PM2.5 Menggunakan LSTM di Daerah Khusus Jakarta

Abstract

Penurunan kualitas udara akibat PM2.5 menjadi masalah kesehatan serius di Daerah Khusus Jakarta yang menempati posisi ke-15 di dunia. Penelitian ini bertujuan mengembangkan kerangka prediksi risiko kesehatan akibat paparan PM2.5 dengan mengintegrasikan data PM2.5, Aerosol Optical Depth (AOD), meteorologi ERA5, dan data Infeksi Saluran Pernapasan Akut (ISPA) dari Sistem Kewaspadaan Dini dan Respon (SKDR) Kementerian Kesehatan Republik Indonesia. Model Long Short-Term Memory (LSTM) digunakan untuk memprediksi konsentrasi PM2.5 dan jumlah kasus ISPA dengan memanfaatkan hubungan yang tidak linear serta pola keterkaitan antarvariabel dari waktu ke waktu, sehingga mampu merepresentasikan kondisi kualitas udara dan risiko kesehatan masyarakat pada periode mendatang. Model forecasting PM2.5 menghasilkan RMSE 4,0397, MAE 3,1963, MAPE 18,5326%, R² 0,7933, model prediksi ISPA menghasilkan RMSE 1912,8461, MAE 1503,7368, MAPE 3,7482%, dan R² 0,9414. Analisis menunjukkan hubungan langsung PM2.5 dan ISPA relatif lemah, namun keterkaitan meningkat hingga 0,215 pada selang waktu 8 minggu yang mengindikasikan adanya dampak yang tidak langsung muncul. Hasil kombinasi prediksi PM2.5 dan prediksi ISPA pada tahun 2026 yaitu didominasi oleh risiko kesehatan kategori sedang. Kerangka yang dikembangkan berpotensi digunakan sebagai early warning system untuk mendukung mitigasi risiko kesehatan akibat polusi udara di Daerah Khusus Jakarta.

The decline in air quality caused by fine particulate matter (PM2.5) has become a major public health concern in the Special Capital Region of Jakarta, which ranks 15th among the world's most polluted cities. This study aimed to develop a health risk prediction framework for PM2.5 exposure by integrating PM2.5 concentration data, Aerosol Optical Depth (AOD), ERA5 meteorological variables, and Acute Respiratory Infection (ARI) data obtained from the Early Warning Alert and Response System (EWARS) of the Ministry of Health of the Republic of Indonesia. A Long Short-Term Memory (LSTM) model was employed to predict PM2.5 concentrations and ARI cases by capturing nonlinear relationships and temporal dependencies among variables, enabling the model to represent future air quality conditions and public health risks. The PM2.5 forecasting model achieved an RMSE of 4,0397, an MAE of 3,1963, a MAPE of 18,5326%, and an R² of 0,7933. Meanwhile, the ARI prediction model achieved an RMSE of 1912,8461, an MAE of 1503,7368, a MAPE of 3,7482%, and an R² of 0,9414. The analysis indicated that the direct relationship between PM2.5 and ARI was relatively weak; however, the correlation increased to 0,215 at an eight-week lag, suggesting a delayed health effect following PM2.5 exposure. The combination of PM2.5 forecasts and ARI predictions for 2026 was dominated by the moderate health risk category. The proposed framework has the potential to serve as an early warning system to support public health risk mitigation associated with air pollution in the Special Capital Region of Jakarta.