Scholarlink

<p>Leptospirosis is a significant zoonotic disease in India, with Gujarat and Karnataka recognised as enzootic states experiencing recurrent animal and human outbreaks. This study applied a spatio-temporal ensemble machine learning framework to monthly animal leptospirosis surveillance data from these regions, integrating satellite-derived environmental predictors. Environmental variables were aligned with surveillance data and processed into continuous monthly time series. Three base regression models, Random Forest (RF), Extreme Gradient Boosting (XGB), and a feed-forward neural network (FFNN) were developed and combined using weighted and stacked meta-ensemble approaches. Model performance was evaluated using Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and the coefficient of determination (R²), and baseline models including linear regression and climatological means were used for comparison. Across both states, the feed-forward neural network performed best among base learners, while the stacked meta-ensemble achieved the lowest prediction errors. However, the near-perfect performance observed (R² ≈ 1.000) should be interpreted cautiously given the limited dataset, passive surveillance data structure, and potential temporal autocorrelation. The models developed in this study therefore forecast reported surveillance cases rather than true disease incidence. SHAP-based feature attribution identified rainfall at short temporal lags as the dominant predictor, with additional contributions from vegetation dynamics. These findings demonstrate the feasibility of integrating environmental predictors with passive surveillance data for short-term forecasting and early warning of leptospirosis risk in data-limited endemic regions. However, the framework should be considered a methodological and exploratory modelling approach rather than a fully operational forecasting system until longer surveillance time series and external validation are available.</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

Spatio-temporal ensemble machine learning forecasts of animal leptospirosis in enzootic Indian States using integrated environmental predictors

  • 期刊论文

  • 发表日期：

    2026年5月27日

查看全文
  • A. Arivazhagan,
  • Vinodhkumar Obli Rajendran,
  • K. Vinodkumar,
  • Keerthi Aaradhana Vinodhkumar,
  • Dharavath Premkumar,
  • Pradeep Chawhan,
  • Noor Saknim Lepcha,
  • Bablu Kumar,
  • K. P. Suresh,
  • B. R. Gulati,
  • V. Balamurugan

摘要

Leptospirosis is a significant zoonotic disease in India, with Gujarat and Karnataka recognised as enzootic states experiencing recurrent animal and human outbreaks. This study applied a spatio-temporal ensemble machine learning framework to monthly animal leptospirosis surveillance data from these regions, integrating satellite-derived environmental predictors. Environmental variables were aligned with surveillance data and processed into continuous monthly time series. Three base regression models, Random Forest (RF), Extreme Gradient Boosting (XGB), and a feed-forward neural network (FFNN) were developed and combined using weighted and stacked meta-ensemble approaches. Model performance was evaluated using Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and the coefficient of determination (R²), and baseline models including linear regression and climatological means were used for comparison. Across both states, the feed-forward neural network performed best among base learners, while the stacked meta-ensemble achieved the lowest prediction errors. However, the near-perfect performance observed (R² ≈ 1.000) should be interpreted cautiously given the limited dataset, passive surveillance data structure, and potential temporal autocorrelation. The models developed in this study therefore forecast reported surveillance cases rather than true disease incidence. SHAP-based feature attribution identified rainfall at short temporal lags as the dominant predictor, with additional contributions from vegetation dynamics. These findings demonstrate the feasibility of integrating environmental predictors with passive surveillance data for short-term forecasting and early warning of leptospirosis risk in data-limited endemic regions. However, the framework should be considered a methodological and exploratory modelling approach rather than a fully operational forecasting system until longer surveillance time series and external validation are available.