Comparative performance of machine learning algorithms for egg size classification in noiler chickens using egg quality traits

This study evaluated the performance metrics of three machine learning algorithms (Support Vector Machine (SVM), Random Forest (RF), and Logistic Regression (LR)) for classifying egg size based on external egg quality traits of Noiler chickens of three plumage varieties. Three hundred freshly laid eggs (100 per plumage variety) were collected at young laying age (26 weeks) and old laying age (46 weeks), and assessed for various quality parameters. The external traits used as features are egg width, egg length, and shape index. Feature importance analysis for RF and LR showed that egg width, and egg length were the most influential predictors of egg size classification, indicating that egg dimensional traits and total egg mass are the primary determinants of egg size in Noiler chickens. Among the models, the RF algorithm achieved the highest overall performance with classification accuracy at (0.87), precision (0.91), recall (0.91), balanced accuracy (0.81) and ROC AUC score (0.93) which indicated exceptional classifying ability. LR recorded classification accuracy at (0.83), precision (1.00), recall (0.78), balanced accuracy (0.89) and ROC AUC score (0.95), while SVM recorded classification accuracy at (0.65), precision (0.86), recall (0.65), balanced accuracy (0.65) and ROC AUC score (0.80) Therefore, the model developed from the RF algorithm can be effectively used for automated egg grading and selection in poultry breeding programs. Future research could incorporate additional features such as computer vision techniques to further enhance classification accuracy.