Trait prediction through computational intelligence and machine learning applied to soybean (Glycine max) breeding in shaded environments

Abstract

This study aims to identify more relevant predictors traits, considering different prediction approaches in soybean under different shading levels in the field, using methodologies based on artificial intelligence and machine learning. The experiments were carried out under different shading levels in a greenhouse and in the field, using sixteen cultivars. We have evaluated grain yield, which was used as a response trait, and 22 other attributes as explanatory traits. Three levels of shading were used to restrict photosynthetically active radiation (RPAR): 0%, 25%, and 48%. At full sun level (0% RPAR), the traits that presented better predictive performances using a multilayer perceptron were specific leaf area, plant height and number of pods. In the three levels of shading, the plant height trait exhibited the best performance for the radial base function network. Plant height showed the best predictive efficiency for grain yield at 25% and 48% RPAR, for all machine learning methodologies. Computational intelligence and machine learning methodologies have proven to be efficient in predicting soybean grain yield, regardless of shading level.