Selection

Each trial solution (individual) is assigned a figure of merit that represents how good a solution it is according to the fitness function (cost function or objective function) of the problem. The most fit individuals (lowest cost-values for minimization problems) are given a higher probability of mating in order to produce the next generation. There are several ways to select the ``parents'' for mating, such as random pairing, roulette wheel, rank weighting and tournament selection Haupt and Haupt (1998). Whatever the selection method, the net effect is to skew the next generation towards the most fit individuals, that is, towards the most promising regions of the search space. It is still possible and desirable to allow less fit individuals to mate, albeit with a lower probability. This increases the exploration of the search space and helps prevent premature convergence, i.e. convergence to a local minimum.