A comparative study of GA and orthogonal experimental design

Our hypothesis is that traditional genetic algorithms (GA) work well because of GA encoding which uses binary variables. GA searching of recombinations or mutations can be replaced by simple statistical methods, orthogonal experimental designs (OED) and factor analysis, while using the same encoding (the same problem representation), and maintaining comparable precision with traditional GA. We describe our development of an orthogonal design algorithm (ODA) for a comparison with GA searching mechanisms. ODA uses GA encoding and OED, but uses no recombinations or mutations. ODA consists of three stages: iterative factor analysis with OED, shuffling, and correction. We compared ODA with traditional GA solving simple quadratic functions and a small protein folding problem, and get comparable results in terms of obtaining approximate solutions.