Artificial intelligence models for dengue outbreak prediction in Ho Chi Minh City, Vietnam under climate change

Dengue fever (DF) remains one of the most significant mosquito-borne diseases in tropical and subtropical regions, particularly in Ho Chi Minh City (HCMC), Vietnam. This study evaluates the performance of Artificial Intelligence (AI) models, including Machine Learning and Deep Learning approaches to forecast weekly district-level DF cases and to identify the optimal predictive model. DF notifications and meteorological data (temperature, humidity, rainfall) from 24 districts in HCMC (2015-2022) were standardized to epidemiological weeks and spatially interpolated using Inverse Distance Weighting (IDW). Variable-specific lags were estimated empirically by Cross-Correlation Analysis (CCA) over 0-21 weeks, and multicollinearity was screened with the Pearson Correlation Coefficient (PCC) and the Variance Inflation Factor (VIF) and resolved with Principal Component Analysis (PCA). Five forecasting models were compared: Multiple Linear Regression (MLR), Support Vector Regression (SVR), Random Forest (RF), Autoregressive Integrated Moving Average (ARIMA), and Long Short-Term Memory (LSTM). Shapley Additive Explanations (SHAP) were applied to assess feature contributions. Strongest lag impacts on DF were found for rainfall (8 weeks), maximum humidity (8 weeks), minimum humidity (8 weeks), minimum temperature (17 weeks), maximum temperature (0 weeks), and temperature amplitude (5 weeks), with correlation |r| ranging from 0.18 to 0.37.Among all models, LSTM achieved the best performance with the lowest errors and highest stability across 2021-2022 test sets. Incorporating lagged meteorological and epidemiological features notably improved predictive accuracy, particularly for MLR. Overall, the LSTM model shows strong potential for dengue forecasting in HCMC and supports the development of AI-based early warning systems for proactive public health management.