Model Selection and Parameter Estimation for Fractional SIR Model Based on the Combination of Reinforcement Learning...

Area of Science:

• Computational epidemiology
• Mathematical modeling
• Machine learning applications

Background:

• Fractional susceptible-infected-recovered (FSIR) models are crucial for understanding disease dynamics.
• Accurate model selection and parameter estimation are vital for reliable epidemiological predictions.
• Traditional methods like Approximate Bayesian Computation (ABC) can be computationally intensive and less effective for complex models.

Purpose of the Study:

• To introduce a novel algorithm that enhances model selection and parameter estimation for FSIR models.
• To integrate reinforcement learning (RL) with approximate Bayesian computation sequential Monte Carlo (ABC-SMC) for improved performance.
• To evaluate the efficacy of the proposed RL-ABC-SMC methodology against existing techniques.

Main Methods:

• Development of a hybrid algorithm combining reinforcement learning (RL) for model selection and ABC-SMC for parameter estimation.
• Implementation of RL to guide the selection process among different FSIR model structures.
• Utilization of ABC-SMC for efficient and accurate parameter inference within the selected model.

Main Results:

• Numerical simulations demonstrate that the RL-ABC-SMC algorithm significantly outperforms the standard ABC-SMC method in model selection accuracy.
• The proposed approach shows enhanced effectiveness in parameter estimation for fractional epidemiological models.
• The RL component effectively directs the search towards more appropriate model structures.

Conclusions:

• The novel RL-ABC-SMC algorithm represents a significant advancement in the analysis of fractional epidemiological models.
• This integrated approach offers a more robust and efficient solution for model selection and parameter estimation challenges.
• The methodology shows promise for real-world applications in disease modeling and public health.