Genetic algorithms (GA's) are search algorithms that imitate nature with th
eir Darwinian survival of the fittest approach. They are well suited for se
arching among a large number of possibilities for solutions because they ex
ploit knowledge contained in a population of initial solutions to generate
new and potentially better solutions. GA's have several advantages over con
ventional search techniques. First, GA's consider many points in the search
space simultaneously. Because GA's utilize parallelism in which a large nu
mber of candidate solutions are simultaneously searched, more of the respon
se surface is probed, so there is a reduced chance of convergence to a loca
l minimum. Second, genetic algorithms make no assumption about the geometry
of the response surface. Hence, discontinuities or singularities in the re
sponse surface, which rule out the use of derivative or simplex based metho
ds, will not pose a problem for GA's. Third, the computational environment
offered by a GA can be readily adjusted to match a particular application.
Thus, GA's can be tailored for individual problems. Consequently, GA's can
be used to solve a variety of data analysis problems in chemistry including
curve fitting, parameter estimation, function optimization. calibration, c
lassification, and wavelength and feature selection.