Determination of milk yield in water buffaloes using multi-class logistic regression and machine learning methods

Area of Science:

• Animal Science
• Machine Learning Applications
• Dairy Production

Background:

• Accurate prediction of milk yield in water buffaloes is crucial for efficient dairy management.
• Machine learning models offer promising tools for analyzing complex biological data and improving predictive accuracy.

Purpose of the Study:

• To comparatively evaluate the predictive performance of Random Forest, Gradient Boosting Machines (GBM), Support Vector Machines (SVM), and Multi-Class Logistic Regression (MCLR) for water buffalo milk yield.
• To identify the most successful machine learning model for milk yield prediction based on key performance metrics.

Main Methods:

• The study employed stratified 8-fold cross-validation and hyperparameter tuning for RF, GBM, and SVM models.
• Data included lactation period, lactation milk yield, age at first pregnancy, and feed type, processed using Python libraries (Pandas, NumPy, Scikit-learn).
• The multicollinear AGE variable was removed to enhance model robustness.

Main Results:

• Gradient Boosting Machines (GBM) demonstrated superior performance, achieving an average accuracy of 64.63%.
• GBM also yielded the highest weighted precision (0.6578), recall (0.6463), F1-score (0.6311), and ROC AUC (0.6625) among the evaluated models.
• While predictive performance was moderate, GBM showed significant potential.

Conclusions:

• The Gradient Boosting Machines (GBM) model is the most effective among the tested algorithms for predicting milk yield in water buffaloes.
• The findings underscore the utility of advanced machine learning techniques in enhancing the precision of dairy animal production predictions.
• Further research and model refinement could lead to improved accuracy and practical applications in water buffalo farming.