Scholarlink

<p>This study introduces a forecasting framework integrating logistic smooth transition (LST) functions with rolling window estimation to predict epidemic peak timing and magnitude. Unlike traditional time series approaches focused on point-by-point accuracy, this methodology provides advance indicators through dynamic parameter recalibration as new observations arrive. Applied to global COVID-19 data, the model successfully identified approaching peaks with 25–90&#xa0;day lead times across diverse epidemic contexts, including detailed validation in China and Germany. The computational architecture (featuring O(T·k·n·i) complexity with parallel structure) scales from single-region analysis on standard hardware to high-performance computing implementations that enable operational multi-region surveillance requiring daily updates across hundreds of regions. The framework transforms point-by-point forecasting into structural parameter estimation, complementing existing epidemiological models while addressing the specific need for early peak detection in public health planning. Performance varies with data quality and intervention impacts, presenting a balance between early warning capability and forecast precision.</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 dynamic estimation framework for epidemic forecasting: the rolling LST model

  • 期刊论文

  • 发表日期：

    2026年5月28日

查看全文
  • Yeliz Uymaz,
  • Tolga Omay

摘要

This study introduces a forecasting framework integrating logistic smooth transition (LST) functions with rolling window estimation to predict epidemic peak timing and magnitude. Unlike traditional time series approaches focused on point-by-point accuracy, this methodology provides advance indicators through dynamic parameter recalibration as new observations arrive. Applied to global COVID-19 data, the model successfully identified approaching peaks with 25–90 day lead times across diverse epidemic contexts, including detailed validation in China and Germany. The computational architecture (featuring O(T·k·n·i) complexity with parallel structure) scales from single-region analysis on standard hardware to high-performance computing implementations that enable operational multi-region surveillance requiring daily updates across hundreds of regions. The framework transforms point-by-point forecasting into structural parameter estimation, complementing existing epidemiological models while addressing the specific need for early peak detection in public health planning. Performance varies with data quality and intervention impacts, presenting a balance between early warning capability and forecast precision.