Linear and non-linear dynamics of the epidemics: System identification based parametric prediction models for the...

Area of Science:

• Epidemiology and Public Health
• Mathematical Modeling
• Infectious Disease Dynamics

Background:

• The COVID-19 pandemic presents significant societal and healthcare challenges, necessitating accurate forecasting for policy development.
• Overwhelmed hospitals require predictive models to manage intensive care and intubation needs.
• Non-pharmacological interventions (NPIs) like lockdowns are crucial for controlling disease spread.

Purpose of the Study:

• To develop and validate mathematical models for predicting future COVID-19 casualties, including intensive care and intubated patients.
• To analyze the linear and non-linear dynamics of the pandemic under various NPIs.
• To compare the performance of machine learning approaches in parameter estimation for these models.

Main Methods:

• Construction of three S p I n I t I b D-N model structures incorporating non-pharmacological policies (N).
• Modification and integration of NPIs into the models, with parameters learned from Turkish Health Ministry data.
• Application of two machine learning techniques: recursive neural networks and batch least squares for parameter optimization.

Main Results:

• The linear S p I n I t I b D-N model demonstrated superior accuracy in predicting COVID-19 casualties under NPIs.
• Non-pharmacological policies were found to have a significant damping effect on pandemic dynamics.
• Batch least squares outperformed recursive neural networks for linear dynamics and stochastic data.

Conclusions:

• The linear S p I n I t I b D-N model provides reliable predictions for COVID-19 trajectories, indicating convergence to zero for suspicious, infected, and deceased cases within 120 days.
• Accurate modeling of NPIs is essential for understanding and mitigating pandemic impacts.
• Machine learning methods offer valuable tools for refining epidemiological models and improving public health strategies.