(344g) Novel Genetic Algorithm for Parameter Fitting Applied to Sigmoid Growth Models

The growth function, also known as the Sigmoid function has applications in a wide variety of engineering and biological applications. This paper presents a novel evolutionary algorithm called Modified Genetic Algorithm (MGA), which combines a multi-generational heuristic search with a region preference ordering to expedite the search. The model has been tested by solving a parameter estimation problem, namely the curve-fitting of a five-parameter Sigmoid model to a set of mortality data from the Covid-19 pandemic.

MGA assigns blocks of eight sets of random parameter values as an initial starting population, and then computes a new generation of models. High-fitness results are maintained in the population from the best of the current generation, and also if they remain in the overall best models up to the current generation. Global diversification (as in a Tabu search algorithm) is implemented if no models in the current generation reach a preset minimum level on the current best solution list. Conversely the search volume is reduced when promising parameters are found, via local intensification. Results compare MGA to traditional curve-fitting algorithms in terms of robustness on the given data sets and computational efficiency.