Scholarlink

<p>Climate-sensitive infections remain a critical threat in tropical coastal settings. We analyzed Ba Ria–Vung Tau (BRVT), Vietnam (2010–2020), to quantify climate–disease relationships and build one-month-ahead forecasts for dengue, acute diarrhea, hand, foot, and mouth disease (HFMD), and influenza. Monthly surveillance counts were linked to satellite/reanalysis climate data and modeled using interpretable statistical approaches (GLMs, DLNMs) alongside machine-learning models (Random Forest, Gradient Boosting) and compact Long Short-Term Memory (LSTM) networks. A rolling-origin time-series cross-validation scheme was combined with an untouched 24-month hold-out set (2019–2020) and leave-one-district-out spatial validation. Temperature variability, humidity, and precipitation were the dominant climatic drivers: a 1&#xa0;°C rise in minimum temperature was associated with a 12.3% increase in dengue incidence (95% CI: 10.1%–14.5%), while a 10% increase in precipitation correlated with a 7.8% rise in acute diarrhea (95% CI: 6.2%–9.4%). LSTM provided the strongest predictive performance (R<sup>2</sup>≈0.62 for dengue; R<sup>2</sup>≈0.43 for HFMD) and ~ 15% lower RMSE than tree-based baselines, whereas performance was modest for influenza (R<sup>2</sup>≈0.33) and acute diarrhea (R<sup>2</sup>≈0.26), indicating substantial non-climatic determinants. District-level risk maps derived from one-month-ahead predictions showed meaningful discrimination and satisfactory calibration for dengue and HFMD on the hold-out period; patterns were directionally consistent under spatial validation. We position these climate-informed AI models as early-warning components for BRVT, emphasizing the need to incorporate sanitation (WASH), vaccination, mobility, and socioeconomic indicators—and to conduct external validation—before broader deployment across regions and under long-term climate scenarios.</p> Graphical Abstract <p></p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

AI-Driven Climate-Health Forecasting: A Machine Learning Model for Epidemic Risk Prediction in Ba Ria-Vung Tau, Vietnam

  • 期刊论文

  • 发表日期：

    2025年11月11日

查看全文
  • Dang Anh Tuan

摘要

Climate-sensitive infections remain a critical threat in tropical coastal settings. We analyzed Ba Ria–Vung Tau (BRVT), Vietnam (2010–2020), to quantify climate–disease relationships and build one-month-ahead forecasts for dengue, acute diarrhea, hand, foot, and mouth disease (HFMD), and influenza. Monthly surveillance counts were linked to satellite/reanalysis climate data and modeled using interpretable statistical approaches (GLMs, DLNMs) alongside machine-learning models (Random Forest, Gradient Boosting) and compact Long Short-Term Memory (LSTM) networks. A rolling-origin time-series cross-validation scheme was combined with an untouched 24-month hold-out set (2019–2020) and leave-one-district-out spatial validation. Temperature variability, humidity, and precipitation were the dominant climatic drivers: a 1 °C rise in minimum temperature was associated with a 12.3% increase in dengue incidence (95% CI: 10.1%–14.5%), while a 10% increase in precipitation correlated with a 7.8% rise in acute diarrhea (95% CI: 6.2%–9.4%). LSTM provided the strongest predictive performance (R2≈0.62 for dengue; R2≈0.43 for HFMD) and ~ 15% lower RMSE than tree-based baselines, whereas performance was modest for influenza (R2≈0.33) and acute diarrhea (R2≈0.26), indicating substantial non-climatic determinants. District-level risk maps derived from one-month-ahead predictions showed meaningful discrimination and satisfactory calibration for dengue and HFMD on the hold-out period; patterns were directionally consistent under spatial validation. We position these climate-informed AI models as early-warning components for BRVT, emphasizing the need to incorporate sanitation (WASH), vaccination, mobility, and socioeconomic indicators—and to conduct external validation—before broader deployment across regions and under long-term climate scenarios.

Graphical Abstract