Scholarlink

<p>This study presents a data-driven framework for modeling the dynamics of a controlled SVIR (Susceptible-vaccinated-infected-recovered) epidemic system using neural ordinary differential equations (Neural ODEs). The proposed approach learns the continuous-time vector field governing the epidemic dynamics directly from data while simultaneously identifying time-dependent vaccination and treatment control functions. Synthetic reference solutions generated via a fourth-order Runge–Kutta (RK4) method are used to benchmark model accuracy. Numerical comparisons demonstrate strong agreement between Neural ODE predictions and classical solutions, with consistently low mean absolute errors across all compartments. Time-series trajectories, error profiles, stacked population distributions, and phase-plane analyses are employed to validate model fidelity and dynamical consistency. The results confirm that Neural ODEs can accurately approximate controlled epidemic dynamics while retaining flexibility, differentiability, and adaptability to time-varying interventions. This framework provides a promising tool for real-time epidemic prediction, intervention assessment, and decision support in dynamically evolving public health and clinical settings.</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

Bridging data and dynamics: neural ODEs for epidemic modeling

  • 期刊论文

  • 发表日期：

    2026年1月22日

查看全文
  • Ahmed M. Zidan,
  • Mohammed Alquraish,
  • Musaad S. Aldhabani,
  • Haitham Qawaqneh,
  • Faheem Jan,
  • Noorullah Noori

摘要

This study presents a data-driven framework for modeling the dynamics of a controlled SVIR (Susceptible-vaccinated-infected-recovered) epidemic system using neural ordinary differential equations (Neural ODEs). The proposed approach learns the continuous-time vector field governing the epidemic dynamics directly from data while simultaneously identifying time-dependent vaccination and treatment control functions. Synthetic reference solutions generated via a fourth-order Runge–Kutta (RK4) method are used to benchmark model accuracy. Numerical comparisons demonstrate strong agreement between Neural ODE predictions and classical solutions, with consistently low mean absolute errors across all compartments. Time-series trajectories, error profiles, stacked population distributions, and phase-plane analyses are employed to validate model fidelity and dynamical consistency. The results confirm that Neural ODEs can accurately approximate controlled epidemic dynamics while retaining flexibility, differentiability, and adaptability to time-varying interventions. This framework provides a promising tool for real-time epidemic prediction, intervention assessment, and decision support in dynamically evolving public health and clinical settings.