Scholarlink

Background <p>Measles has re-emerged as a growing public health threat despite the availability of a safe and effective vaccine. In Bangladesh, recurring outbreaks and widening immunization disparities underscore the urgent need for proactive district-level risk assessment. This study developed an explainable machine learning framework to identify drivers of and forecast district-level measles outbreak risk in Bangladesh.</p> Methods <p>District-level measles surveillance data (2017–2023) were combined with climatic, socio-demographic, healthcare capacity, and vaccination indicators. Eight machine learning classifiers (Decision Tree, Support Vector Machine, Naive Bayes, Random Forest, LightGBM, CatBoost, XGBoost, and an Elastic Net GLM) were compared using temporal train /test splitting, and nested cross-validation with F1-optimised decision thresholds selected within training fold. XGBoost was selected based on superior AUC-ROC and Matthews Correlation Coefficient. SHapley Additive exPlanations quantified predictor contributions on measles outbreak prediction. The model was retrained on the full 2017–2023 dataset to forecast district-level outbreak probabilities for 2024–2028, with 95% confidence intervals.</p> Results <p>Measles showed marked spatiotemporal heterogeneity shifting from northeastern districts during 2017–2018 toward the Dhaka periphery during 2021–2023. XGBoost achieved an AUC-ROC of 0.838 (accuracy 81.3%; F1 0.840; MCC 0.624), supporting risk-based district prioritisation rather than precise case-level prediction. Sociodemographic factors contributed most to the outbreak prediction(41.37%), followed by climatic(39.22%) and healthcare-related variables(19.41%). Vaccination coverage showed lower standalone importance, reflecting contextual risk modulation. Minimum temperature, household size, and poverty headcount were the dominant predictors of measles outbreak prediction. Sylhet, Kushtia, and Dhaka were most vulnerable districts of measles outbreak, while Khulna and Chittagong divisions maintained persistently elevated risk through 2028.</p> Conclusion <p>Measles outbreak risk in Bangladesh is shaped by climatic and socioeconomic vulnerabilities alongside vaccination gaps. Explainable machine learning can support precision immunisation and targeted surveillance, while requiring cautious interpretation of forecast uncertainty.</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

District-level measles outbreak prediction in Bangladesh using geospatial artificial intelligence

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年8月4日

查看全文
  • Md. Abu Bokkor Shiddik,
  • Md. Siddikur Rahman

摘要

Background

Measles has re-emerged as a growing public health threat despite the availability of a safe and effective vaccine. In Bangladesh, recurring outbreaks and widening immunization disparities underscore the urgent need for proactive district-level risk assessment. This study developed an explainable machine learning framework to identify drivers of and forecast district-level measles outbreak risk in Bangladesh.

Methods

District-level measles surveillance data (2017–2023) were combined with climatic, socio-demographic, healthcare capacity, and vaccination indicators. Eight machine learning classifiers (Decision Tree, Support Vector Machine, Naive Bayes, Random Forest, LightGBM, CatBoost, XGBoost, and an Elastic Net GLM) were compared using temporal train /test splitting, and nested cross-validation with F1-optimised decision thresholds selected within training fold. XGBoost was selected based on superior AUC-ROC and Matthews Correlation Coefficient. SHapley Additive exPlanations quantified predictor contributions on measles outbreak prediction. The model was retrained on the full 2017–2023 dataset to forecast district-level outbreak probabilities for 2024–2028, with 95% confidence intervals.

Results

Measles showed marked spatiotemporal heterogeneity shifting from northeastern districts during 2017–2018 toward the Dhaka periphery during 2021–2023. XGBoost achieved an AUC-ROC of 0.838 (accuracy 81.3%; F1 0.840; MCC 0.624), supporting risk-based district prioritisation rather than precise case-level prediction. Sociodemographic factors contributed most to the outbreak prediction(41.37%), followed by climatic(39.22%) and healthcare-related variables(19.41%). Vaccination coverage showed lower standalone importance, reflecting contextual risk modulation. Minimum temperature, household size, and poverty headcount were the dominant predictors of measles outbreak prediction. Sylhet, Kushtia, and Dhaka were most vulnerable districts of measles outbreak, while Khulna and Chittagong divisions maintained persistently elevated risk through 2028.

Conclusion

Measles outbreak risk in Bangladesh is shaped by climatic and socioeconomic vulnerabilities alongside vaccination gaps. Explainable machine learning can support precision immunisation and targeted surveillance, while requiring cautious interpretation of forecast uncertainty.