Seleção de famílias de cana-de-açúcar via árvores de decisão

• Autor(a) principal: Bernardes, Diego Paiva
• Data de Publicação: 2013
• Tipo de documento: Dissertação
• Idioma: por
• Título da fonte: LOCUS Repositório Institucional da UFV
• Texto Completo: http://locus.ufv.br/handle/123456789/4069
• topic: CART; Melhoramento; Árvore de regressão; CART; Improvement; Regression tree; CNPQ::CIENCIAS AGRARIAS

The process of selection of clones to be used as new commercial varieties is lacking in statistical methods and phytotechnical that increase the genetic gains in crop improvement programs of cane sugar. Out of five stages of the selection program to improve the RIDESA, the first one, T1, is endowed with great importance. First because it develops the remaining phases of the crop breeding program. And second because the volume of information to be analyzed is huge. Thus, if the data is not properly analyzed, good clones can be ruled out in the early stages, reducing the excellence of the program. The common strategies of selection, BLUP and BLUPIS, have the disadvantage of the necessity of weighing the whole plot. One way around this is to categorize the components of production; stalk height, stalk diameter and number of stalks, by decision trees. Through these trees, you can generate the combinations of these yield components and their production values. Using data from commercial varieties to generate the trees, it would not be necessary to weigh the entire plot, saving time and money. The aim of this study was to evaluate the categorization of yield components as selection strategy between and within families by comparing their performance with the usual methods, BLUP and BLUPIS. The algorithm used was the CART. Nonparametric by nature, it`s capable of producing binary divisions combining the explanatory variables and associating them with different response values. Data were collected from 5 experiments, installed in May 2007 in a randomized block design, with each experiment consisting of 5 blocks, 22 families and two commercial varieties. CART algorithm was effective in defining classes of yield components followed by selection of the best families with mean accuracy of 73% when compared with BLUPIS and BLUP.