Supersaturated experimental designs: new approaches to building and using it Part II. Solving supersaturated designs by genetic algorithms

A new procedure for the resolution of supersaturated matrices is presented. This procedure is based on genetic algorithm-driven regression, using an e fficient system of genetic operators' competence. Genetic regression is com pared with stepwise regression and all-subsets regression that has been pro posed to solve this type of matrices, presented results show that this new genetic-driven procedure outperforms both stepwise and all-subsets regressi on procedures. The proposed method shows satisfactory efficiency and robust ness for data even in the presence of gross accidental errors. All studies were conducted by a dedicated toolkit that is briefly described. (C) 2001 E lsevier Science B.V. All rights reserved.