Artificial neural networks and linear discriminant analysis in early selection among sugarcane families

Abstract

One of the major challenges in sugarcane breeding programs is an efficient selection of genotypes in the initial phase. The purpose of this study was to compare modelling by artificial neural networks (ANN) and linear discriminant analysis (LDA) as alternatives for the selection of promising sugarcane families based on the indirect traits number of sugarcane stalks (NS), stalk diameter (SD) and stalk height (SH). The analysis focused on two models, a full one with all predictors, and a reduced one, from which the variable SH was excluded. To compare and assess the applied methods, the apparent error rate (AER) and true positive rate (TPR) were used, derived from the confusion matrix. Modeling with ANN and LDA can be used successfully for selection among sugarcane families. The reduced model may be preferable, for having a low AER, high TPR and being easier to obtain in operational terms.