Scholarlink

<p>Accurate forecasting of infectious diseases drives modern public health interventions that reduce morbidity and mortality. However, accurate forecasting in real-time remains a challenge for the modeling community. Ensembling has emerged as a critical tool for accurate forecasting by leveraging multiple component (individual) models into a single weighted average. Traditional ensembling strategies have relied on bespoke component models that weight the contributions of individual models according to extensive historical data for specific diseases. This is impractical for an emerging disease, since there would be very little – if any – data. We propose an ensembling strategy, called <i>epiFFORMA</i>, that determines component weights for an ensemble model without historical data and is therefore disease-agnostic. The epiFFORMA model builds upon the FFORMA model from the M4 forecasting competition to harness epidemiological dynamics through synthetic data. We demonstrate that epiFFORMA performs better than a naive, equal-weighting ensembling strategy when forecasting outbreaks of COVID-19, diphtheria, influenza-like illness, dengue, measles, mumps, polio, rubella, smallpox, and chikungunya. We further show that epiFFORMA, on average, performs better than the individual component models in the ensemble.</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

A disease-agnostic approach to ensemble learning for infectious disease forecasting

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年3月20日

查看全文
  • Alexander C. Murph,
  • Lauren J. Beesley,
  • G. Casey Gibson,
  • Lauren A. Castro,
  • Sara Y. Del Valle,
  • Dave Osthus

摘要

Accurate forecasting of infectious diseases drives modern public health interventions that reduce morbidity and mortality. However, accurate forecasting in real-time remains a challenge for the modeling community. Ensembling has emerged as a critical tool for accurate forecasting by leveraging multiple component (individual) models into a single weighted average. Traditional ensembling strategies have relied on bespoke component models that weight the contributions of individual models according to extensive historical data for specific diseases. This is impractical for an emerging disease, since there would be very little – if any – data. We propose an ensembling strategy, called epiFFORMA, that determines component weights for an ensemble model without historical data and is therefore disease-agnostic. The epiFFORMA model builds upon the FFORMA model from the M4 forecasting competition to harness epidemiological dynamics through synthetic data. We demonstrate that epiFFORMA performs better than a naive, equal-weighting ensembling strategy when forecasting outbreaks of COVID-19, diphtheria, influenza-like illness, dengue, measles, mumps, polio, rubella, smallpox, and chikungunya. We further show that epiFFORMA, on average, performs better than the individual component models in the ensemble.