Selection index as a priori information for using artificial neural networks to classify alfalfa genotypes.

• Autor(a) principal: SANTOS, I. G. dos
• Data de Publicação: 2019
• Outros Autores: CRUZ, C. D., NASCIMENTO, M., FERREIRA, R. de P.
• Tipo de documento: Artigo
• Idioma: eng
• Título da fonte: Repositório Institucional da EMBRAPA (Repository Open Access to Scientific Information from EMBRAPA - Alice)
• Texto Completo: http://www.alice.cnptia.embrapa.br/alice/handle/doc/1109207
• topic: Tai index; Computational intelligence; Medicago Sativa

The efficiency of a selection index generally depends on the quality of the variance matrixes, which demands controlled experiments. Using Artificial Neural Networks (ANNs) trained from a selection index is advantageous for selecting genotypes since an ANN has the capacity to classify genotypes in an automated way. We propose the use of ANNs for the selection of alfalfa genotypes, based on a selection index. Data were collected from 77 alfalfa genotypes evaluated based on nine traits from four cuttings. The traits were divided into forage yield and nutritive value groups. In order for the ANNs to learn the classification pattern, the Tai index was used, which allows secondary traits to be included in the index to improve the gains of the main traits. An index was established for each group of traits, and based on the index scores the genotypes were subdivided into four classes (optimal, good, medium, and bad). After testing different topologies, ANNs were established for each index, according to the apparent error rates. The chosen ANNs were efficient in classifying the genotypes since the highest apparent error rate reached 15%, meaning that the ANNs efficiently captured the data pattern. Considering the ANN classification for both groups of traits, there was a high degree of agreement with the classification obtained from the Tai index, as expected. Even in the cuttings where the ANNs presented the worst performance, their potential to classify alfalfa genotypes was clear, because the wrong classifications were placed in groups close to the correct ones. This ensured that the best genotypes did not run the risk of being discarded, since they would not classified in the group of bad genotypes. The ANNs that were developed have good potential for use in alfalfa breeding programs.