Application of a new genetic operator in feature selection problems

Crossover is an important genetic operation that helps in random recombination of structured information to locate new points in the search space, in order to achieve a good solution to an optimization problem. The conventional crossover operation when applied on a pair of binary strings will usually not retain the total number of 1s in the off-springs to be the same as that of their parents. But there are many optimization problems which require such a constraint. We propose a new crossover technique called, "self-crossover", which satisfies this constraint as well as retains the stochastic and evolutionary characteristics of genetic algorithms. As an illustration, the effectiveness of this new technique has been demonstrated for the feature selection problem of pattern recognition.