Estimation of Egg Albumen Height of Hy-Line Hens Using Ridge Regression Method

Abstract

This study aims to predict egg albumen height, one of the internal quality parameters of eggs from Hy-Line hens, using external quality traits. In the study, a high level of multicollinearity problem was detected in the multiple linear regression analysis based on the classical Least Squares Method (LSM).This negatively impacted the predictive power of the model and the reliability of the parameter estimates. To address this structural problem, the Ridge regression method was applied, which adds a penalty term to the regression coefficients and increases the model's stability by reducing the variance. In the analyses, external quality data (weight, width, length, shape index, and Haugh unit) from 150 eggs obtained from 53-week-old Hy-Line hybrid hens were used. Egg albumen height, which has a high coefficient of determination, was selected as the dependent variable. Although the R² value in the model generated using the Least Squares Method (LSM)was 99 %, the high VIF values indicated serious multicollinearity problem. Ridge regression analyses showed that VIF values decreased to acceptable levels, the model's predictive accuracy increased, and the parameters became more statistically significant. In conclusion, it has been demonstrated that multicollinearity problem, frequently encountered in biological data, can be effectively addressed through regularization methods such as Ridge regression. This method is considered an important tool for developing more reliable and consistent models, particularly in selection and genetic breeding studies.