Spatial Distribution Analysis and Comparative Forecasting of Dengue Resurgence in the Philippines (2025–2027): A Nationwide Study

Abstract

Prediction of dengue continues to be valuable in endemic countries. Time series forecasting methods have been widely employed for predicting future dengue trends and outbreaks. The study aimed to determine the spatial distribution, trends, and seasonality of dengue cases and compare the predictive accuracy of seasonal autoregressive integrated moving average (SARIMA), neural network autoregression (NNAR), random forest (RF), long–short term memory (LSTM), trigonometric exponential smoothing state–space model with Box–Cox transformation, ARMA errors, trend and seasonal components (TBATS), and Prophet in forecasting dengue cases in the Philippines. Monthly data from 2017 to 2024 across all provinces were obtained and were partitioned into training (January 2017–December 2023) and testing segments (January 2024–December 2024). Model performance was assessed by analyzing the training data using time series techniques and comparing the resulting forecasts with empirical values from the test dataset. In total, 3-year projections were generated by implementing the models on the entire dataset. The study analyzed 1,903,425 dengue cases with a mean monthly incidence of 17.66 ± 15.97 per 100,000 population. Regular seasonal epidemics were identified, peaking from July to September. NNAR outperformed the other models and predicted an annual average of 444,678 cases from 2025 to 2027. This is the first study to apply SARIMA, RF, LSTM, TBATS, and Prophet in forecasting dengue cases in the Philippines at a national scale. The study offers new insights into disease forecasting, particularly in the application of advanced time series methodologies. These findings should be considered to strengthen surveillance, prevention, and control against dengue.