A Comparative Study on COVID-19 Dynamics: Mathematical Modeling, Predictions, and Resource Allocation Strategies in...

Area of Science:

• Epidemiology
• Mathematical Biology
• Data Science

Background:

• The COVID-19 pandemic highlighted the need for robust predictive models.
• Real-world data often presents inconsistencies and heterogeneity across different regions.
• Existing models may struggle with generalizability due to varying public health interventions and initial conditions.

Purpose of the Study:

• To develop and validate a transferable mathematical modeling framework for infectious disease dynamics.
• To assess the generalizability of a COVID-19 model across heterogeneous national contexts.
• To integrate neural networks and ARMAX models for enhanced data-driven epidemic prediction.

Main Methods:

• A multiple-input single-output (MISO) mathematical model was developed for COVID-19 dynamics.
• Neural networks were used for pre-processing and training model inputs to handle data inconsistencies.
• An Autoregressive Moving Average with eXogenous inputs (ARMAX) model captured the epidemic's stochastic elements.
• The model, initially trained on Romanian data, was applied to data from Italy and Switzerland.

Main Results:

• The Romanian-developed model achieved high forecast accuracy (>98.59%) across Italy and Switzerland.
• The model demonstrated robust performance despite significant differences in testing, policies, and initial case timing.
• The unified data-driven framework proved transferable to heterogeneous national settings.
• Integration of neural networks and ARMAX modeling improved predictive capabilities.

Conclusions:

• A unified, data-driven mathematical modeling framework can be effectively transferred across diverse national contexts.
• Mathematical modeling, enhanced by predictive analytics, offers a powerful tool for understanding and managing global health crises.
• This approach strengthens preparedness for future pandemics by enabling evidence-based decision-making.