Differentiation and classification of productive efficiency of chicken farms using logistic regression and linear discriminant analysis

Abstract

Objective: This study was carried out to compare the accuracy of linear discriminant analysis (LDA) and binary logistic regression (BLR) in classification of level of production in layers (high and low) as a dependent variable using breed, total ration, number of mortality, marketing weight and marketing age as independent variables. Regarding the assumptions of each method, LDA and BLR were also compared with respect to the effect of sample size with consecration to lack of multivariate normality of predictors. Procedures: Record data of 12500 layers were collected from private farms in Dakahlia Governorate during the period from 2018 to 2020). The comparison between LDA and BLR based on the significance of coefficients, classification rate, and areas under ROC curve (AUC). Results: showed that both methods selected breed, total ration consumed and marketing age as significant predictors (P < 0.01) for classification process. The percentages of correct classification for LDA and BLR were 67.7% and 88.9%, respectively. The AUCs were 0.682 and 0.734, for LDA and BLR, respectively. In addition, the sample size effect had the same impact on both analyses, whereas the accuracy of correctly classified cases was higher in BLR than LDA. Conclusion: It could therefore be concluded that LDA and BLR can be used effectively for classification and prediction of level of production in layers even with the lack of normality assumption.