| Genetic algorithms, a popular technique for optimization, traditionally uses binary strings to represent problem parameters and uses crossover and mutation operators based on this binary string representation. However, crossover operations on binary strings result in a highly biased set of possible outcomes, causing some values to have selective pressure separate from the fitness functions representing the optimization problem. This effect is particularly problematic for categorical variables where no notion of genetic similarity applies, thus no fitness selective pressure will be applied to make up for the bias against some values. For this reason, binary strings should not be used in genetic algorithms, especially for categorical variables. |
