Akademik Veri Yönetim Sistemi
Livestock Science, cilt.308, 2026 (SCI-Expanded, Scopus)
Aim of the study was to predict sensory meat quality traits in goat kids using Decision Tree (DT), AdaBoost (AB), Gradient Boosting (GB), Random Forest (RF), K-Nearest Neighbors (KNN), Artificial Neural Networks (ANN), Support Vector Machine (SVR), Lasso, Ridge, Elastic Net and Partial Least Squares (PLS) Regression analysis. The dataset comprised 118 goat kids. Animal-related and rearing factors, carcass characteristics, and instrumental meat quality variables were used as predictors. In the prediction of tenderness, PLS, Lasso, Elastic Net regressions, and SVR were determined as the top-performing algorithms, with R2 values of 0.754, 0.736, 0.735, and 0.736, respectively. The two models that achieved the highest predictive performance for juiciness in terms of R2 were the GB and Ridge regression models, with R2 values of 0.415 and 0.347, respectively. The highest prediction accuracy for odour and flavour intensity was achieved using the SVR and PLS regression algorithms, with R2 values of 0.548 and 0.598 for odour, and 0.550 and 0.478 for flavour intensity, respectively. Warner-Bratzler shear force value was the most important feature for tenderness and juiciness. According to the SVR and PLS regression, cooking loss and slaughter age were the key predictors of odour, whereas pH0–24 h and longissimus thoracis muscle section area were the most influential predictors of flavour intensity. In conclusion, sensory tenderness can be predicted with considerable accuracy using machine learning with animal-related and rearing factors, carcass characteristics, and instrumental meat quality variables, while the prediction of juiciness, odour and flavour intensity was achieved with low to moderate accuracy.