Öznitelik Seçimi Için Çoklu-Ebeveyn Çaprazlama Operatörlerinin Karsilastirilmasi

Genetic algorithms (GA) are mostly used for feature selection in literature. In this study, multi-parent crossover operators are used in genetic algorithms for feature selection. The uniform crossover, occurrence-based crossover, fitness-based crossover, and diagonal crossover are considered as multi-parent crossover operators. Binary representation is used for encoding a candidate solution in GA. The empirical evaluation of these operators is performed on three different datasets with different numbers of features. We perform Oneway ANOVA and Tukey HSD tests at a confidence level of 95 % for statistical comparisons of algorithms. The experiments are conducted into two parts: (1) In this part, we investigate the effect of the number of parents for each multi-parent crossover operator, (2) we compare the performance of these crossover operators with the best setting. The results reveal that occurrence-based crossover with 5 parents outperforms the other crossover operators, however it selects more attributes.