Genetic clustering based on segregation distortion caused by selfish genes

摘要

In this study we propose a data clustering method relying on a new genetic operator, which is based on the biological concept of segregation distortion genes. The proposed method repeatedly applies conditional recombination and mutation operators to a pair of randomly selected chromosomes from a population, whose initial members represent the target data to be clustered, thus increasing the population size. While doing so, so-called segregation distortion genes are recognized, which then separate the growing population into species. There, a species is characterized by a set of chromosomes that can yield new chromosomes by using standard genetic operators, while these operators cannot be applied between the set of chromosomes of different species (this way establishing above-mentioned conditional application of operators). This also indicates that the proportion of a particular allele on some locus within the whole population, in comparison to other alleles, can increase, thus giving raise for new segregation distortion genes, and new species, to appear. The assignment of the initial cluster data within the population to species gives the clustering result. The proposed method is demonstrated for the problem of clustering of bit strings, the processing is analyzed, and its feasibility is shown.