Construction of exact D-optimal designs for linear regression models using genetic algorithms

Abstract

This paper reports the application of genetic algorithms to the construction of exact D-optimal designs for linear regression models. D-optimal designs and the 2-exchange Federov algorithm are introduced. A modification of this algorithm, based on simulated annealing, is then discussed. This procedure is taken as a basis for the cross-over and mutation operators used by the genetic algorithm that is proposed in this paper. Finally, the performance of this algorithm is illustrated by applying it to the construction of exact D-optimal designs for the 2nd order model with 2 factors on the domain [-1.0, 1.0] x [-1.0, 1.0].