Composite clustering crossover operator for improving genetic algorithms

Introduced by Holland in 1975, genetic algorithms are used to search for solutions in cases where the search area is big. Standard implementations of genetic algorithms use well known operators like parent selection, crossover and mutation. For each type of operators, there are different possible implementations to choose from. In most of the cases, however, the crossover implementations are variations of the same idea, concentrated around cutting parent chromosomes into parts and using these parts to build offspring. There are few exceptions that can be found in literature related to k-means clustering crossover. This paper presents an additional insight into the clustering crossover by analyzing the results of binary, integer and real coded genetic algorithms (RGAs). The experimental work compares a standard genetic algorithm solution for common RGA benchmark functions, OneMax problem as binary coded GA and an integer coded RNA representation, with an original implementation which uses a clustering crossover operator. The results show that the usage of the composite clustering crossover operator can increase the performance of the genetic algorithm in some cases, leading to faster convergence to the desired solution. However, not all problem types benefit from the combination. Further study is required.