Graph Theory Based Large-Scale Machine Learning With Multi-Dimensional Constrained Optimization Approaches for Exact...

Area of Science:

• Epidemiology
• Mathematical Modeling
• Computational Intelligence

Background:

• Pandemic disease prediction models require accurate parameter estimation.
• Existing models may not fully capture complex epidemiological dynamics and constraints.

Purpose of the Study:

• To develop and evaluate graph theory-based constrained multi-dimensional (CM) mathematical and meta-heuristic algorithms (MA) for learning parameters in large-scale epidemiological models.
• To address challenges posed by parameter sign and magnitude constraints, and coupling parameters within sub-models.

Main Methods:

• Construction of a graph theory-based constrained multi-dimensional (CM) mathematical framework.
• Development of gradient-based CM recursive least square (CM-RLS) algorithm.
• Implementation of three search-based meta-heuristic algorithms: CM particle swarm optimization (CM-PSO), CM success history-based adaptive differential evolution (CM-SHADE), and CM-SHADEWO (CM-SHADE enriched with whale optimization).

Main Results:

• CM-RLS demonstrated superior performance when gradient information was available.
• CM-SHADEWO effectively handled hard constraints, uncertainties, and lack of gradient information, capturing dominant CM optimization characteristics.
• All developed algorithms provided satisfactory parameter estimates under varying conditions.

Conclusions:

• Constrained multi-dimensional models and meta-heuristic algorithms offer robust approaches for epidemiological parameter learning.
• CM-SHADEWO is a promising method for complex epidemiological modeling scenarios lacking gradient information.
• The study highlights the importance of tailored optimization techniques for accurate pandemic prediction.