Predicting Malaria Incident Using Hybrid SARIMA-LSTM Model

Abstract

Malaria remains a significant global health concern, particularly in regions with high transmission rates. Accurate and timely prediction of malaria incidence can assist health authorities and policymakers in implementing effective prevention and control measures. However, because data are in limited supply, most of the relevant research studies concentrated on monthly or quarterly data. This study proposes a hybrid forecasting model combining Seasonal Autoregressive Integrated Moving Average (SARIMA) and Long Short-Term Memory (LSTM) neural networks to predict malaria incidence. The hybrid approach enhances accuracy and robustness by capturing historical data’s temporal dependencies and seasonal patterns. The methodology involves collecting historical malaria incidence data, preprocessing it, fitting SARIMA models, extracting residuals, and training LSTM neural networks on residuals. These models capture nonlinear and complex data components, making accurate predictions and capturing long-term dependencies. After training, the hybrid SARIMA-LSTM model is created by combining the predictions from both models. This integration ensures that both the temporal and nonlinear patterns are considered, leading to improved forecast accuracy. Finally, the model is evaluated using appropriate performance metrics, such as mean absolute percentage error (MAPE) or root mean square error (RMSE). The hybrid SARIMA-LSTM model outperforms SARIMA and LSTM in predicting malaria incidence and its accuracy was evaluated through comparisons with other forecasting methods. It captures temporal and nonlinear patterns, enabling timely resource allocation, intervention planning, and proactive measures for improved control and prevention efforts.