Identification of maize (Zea mays L.) progeny genotypes based on two probabilistic approaches: Logistic regression and naïve Bayes

Abstract

We used two probabilistic methods, Gaussian Naïve Bayes and Logistic Regression to predict the genotypes of the offspring of two maize strains, the BLC and the JNE genotypes, based on the phenotypic traits of the parents. We determined the prediction performance of the two models with the overall accuracy and the area under the receiver operating curve (AUC). The overall accuracy for both models ranged between 82% and 87%. The values of the area under the receiver operating curve were 0.90 or higher for Logistic Regression models, and 0.85 or higher for Gaussian Naïve Bayes models. These statistics indicated that the two models were very effective in predicting the genotypes of the offspring. Furthermore, both models predicted the BLC genotype with higher accuracy than they did the JNE genotype. The BLC genotype appeared more homogeneous and more predictable. A Chi-square test for the homogeneity of the confusion matrices showed that in all cases the two models produced similar prediction results. That finding was in line with the assertion by Mitchell (2010) who theoretically showed that the two models are essentially the same. With logistic regression, each subset of the original data or its corresponding principal components produced exactly the same prediction results. The AUC value may be viewed as a criterion for parent-offspring resemblance for each set of phenotypic traits considered in the analysis.