Selection of a Complex of Informative Features for Assessing the Internal Characteristics of Eggs for Consumption

This paper concerns the methods and algorithms developed for tracking changes in the internal characteristics of eggs during storage that emphasize rapid and simple classification and satisfactory regression accuracy, considering technological requirements. A thorough analysis was carried out for 53 characteristics of yolk, thick albumen, and thin albumen, followed by feature selection using the most informative features. Among others, three feature selection methods (FSRNCA, SFCPP, and RReliefF) gave comparably good results and allowed for the identification of the most important features. The results show that with features selected from data across three different manufacturers of hen eggs, more features gave positive cross-validation results and normal distribution checks in each storage stage. This infers that yolk characteristics are more stable and predictable than albumen characteristics. The study also identifies that the numbers and informativeness of selected features of yolk and albumen for hen eggs exceed those for quail eggs, including for cross-validation. Quail eggs give very varied results depending on the producer. Over 60% of the selected albumen features for manufacturers M1 and M2 showed strong performance in both normal distribution checks and cross-validation at different storage stages, while for producer M3, only 17% of features met these criteria. Among the yolk features selected for producers M2 and M3, more than half showed positive outcomes from normal distribution to cross-validation. For producer M1, only 4% of the yolk features passed these criteria at all storage stages. These results highlight the feature stability and predictability variation across different types of eggs and their manufacturers.