Computational intelligence to study the importance of predictors in white oat (Avena sativa L.)

Abstract

The objective of this work was to estimate the best approach for prediction and establish a network with better predictive power in white oat using methodologies based on regression, artificial intelligence, and machine learning. Seventy-eight white oat genotypes were evaluated in 2008 and 2009. Were evaluated without and with fungicide, established prediction models in four experimental sets. The characteristics evaluated were grain yield, which was used as a response variable, and ten others as explanatory variables. Assessing the importance of variables through the impact of destructuring or disturbing the information of a given input on the estimation of R2. This importance was estimated by exchanging information or making the phenotypic value of each characteristic constant and checking for changes in the estimates of R2. When the values of a feature are disturbed, the value of R2 decreases, indicating that the feature is important over the others for prediction purposes. The importance of variables using the radial basis function network was estimated according to the MLP. For machine learning, decision trees, bagging, random forest, and boosting were used. The quality of the predictive model was adjusted based on R2 was used to quantify the importance of the phenotypic trait. The characters indicated to assist in decision-making are plant height, leaf rust severity, and lodging percentage. The R2 ranged from 30.14% − 96.45% and 10.57% − 94.61%, for computational intelligence and machine learning, respectively. The bagging technique showed a high estimate of the coefficient of determination more elevated than the others.