Genetic algorithms in parameter estimation for nonlinear regression models: an experimental approach

Abstract

In this study, we examine the genetic algorithms (GAs) for parameter estimation of nonlinear regression models over a large set of test problems with three difficulty levels. A GA is developed based on a full-factorial experimental design. The proposed GA performs well for the test problems even for relatively larger intervals of the parameters. We analyze the effect of the preset difficulty levels of the test problems on the performance and the convergence properties of the developed GA. On the basis of the convergence analysis, we point out that the regression parameters that have stronger effect on the sum of squared errors converge much faster to their optimal values.