start prev index next end Slide 12 of 18.


Often the objective scores must be transformed in order to help the genetic algorithm maintain diversity or differentiate between very similar individuals. The transformation from raw objective scores to scaled fitness scores is called scaling.

There are many different scaling algorithms. Some of the most common are linear (fitness proportionate) scaling, sigma truncation scaling, and sharing. Linear scaling transforms the objective score based on a linear relationship using the maximum and minimum scores in the population as the transformation metric. Sigma truncation scaling uses the population's standard deviation to do a similar transformation, but it truncates to zero the poor performers. Sharing derates the score of individuals that are similar to other individuals in the population.

For a complete description of each of these methods, see Goldberg's book.

Return to Matthew's Portfolio