The behavior of components of protein plant is of vital importance for animals that consume them in their diet. The objective of this research is to evaluate regression algorithms, to determine the behavior of the expressions that best adapt to the procedures of a traditional laboratory and to estimate the chemical components of protein plants, in this sense the MULAN library of java has been used, that contain automatic learning algorithms capable of adapting to dissimilar problems. Three data set were created for each species treated in this study; each of these include the main elements to be evaluate in each experiment, these are delimits by: secondary metabolites, cell wall components and digestibility element for training files one, two and three, respectively; subsequently, they were evaluated through learning supervised and cross-validation of each to determine the best fit by aRMSE (Average Root Mean Square Error). The learning results were compare with previous experiments, where there was a learning variant that contained in a single dataset all the components to be evaluates in a single prediction. The result of the comparison shows that the lazy algorithms based on instances have a better learning behavior than the others evaluate.