Predicting Malaria Incidence Using LSTM and Environmental Variables

Sulistijanti, Wellie (Institut Teknologi Statistika dan Bisnis Muhammadiyah Semarang)
Khikmah, Laelatul (Institut Teknologi Statistika dan Bisnis Muhammadiyah Semarang)
Rahmasari, Erisa Adyati (Universitas Dian Nuswantoro)
Sangnandha, Cikal Arbitan Putra (Institut Teknologi Statistika dan Bisnis Muhammadiyah Semarang)
Yusuf, Idan Maulana (Institut Teknologi Statistika dan Bisnis Muhammadiyah Semarang)
Azizah, Dzahari Alikharimah (Institut Teknologi Statistika dan Bisnis Muhammadiyah Semarang)

Zero : Jurnal Sains, Matematika, dan Terapan
Vol 9, No 2 (2025): Zero: Jurnal Sains Matematika dan Terapan

Climate change is exacerbating malaria risk in Indonesia, especially in Papua. This study proposes a Bidirectional Long Short-Term Memory (LSTM) model to forecast malaria incidence using climate variables. The dataset comprises monthly malaria and climate records (rainfall, temperature, humidity) from four high-endemic provinces between 2014 and 2024. Key methodologies included data augmentation to address data imbalances and a grouped time-series cross-validation for robust model evaluation. An ARIMA model was implemented as a validation baseline to benchmark the proposed approach. The Bi-LSTM model delivered superior performance, achieving an average test R² of 0.7210 and SMAPE of 11.02%. the model demonstrated excellent generalization with no evidence of overfitting, significantly outperforming the ARIMA baseline. The findings validate the use of deep learning models as effective tools for public health surveillance, providing reliable early warnings to support timely interventions. Future work will apply SHAP interpretability techniques and expanding the model's geographic scope.