Optimized Machine Learning Models for Predicting Core Body Temperature in Dairy Cows: Enhancing Accuracy and...

Area of Science:

• Animal Science
• Machine Learning
• Environmental Physiology

Background:

• Heat stress significantly impacts dairy cow health and productivity.
• Effective management strategies are crucial for animal welfare and farm economics.
• Predictive modeling offers a novel approach to proactive heat stress mitigation.

Purpose of the Study:

• To develop and validate a machine learning framework for predicting dairy cow core body temperature (CBT).
• To identify key physiological and environmental factors influencing CBT.
• To optimize predictive model performance for enhanced heat stress management.

Main Methods:

• Utilized a dataset of 3005 physiological records from dairy cows in production environments.
• Applied various machine learning algorithms including Elastic Net, Artificial Neural Networks, Random Forests, XGBoost, LightGBM, and CatBoost.
• Employed Bayesian Optimization and Grey Wolf Optimizer for hyperparameter tuning and model refinement.

Main Results:

• The feature set including average infrared trunk temperature (IRTave_TK) showed strong predictive capability (R²=0.516, MAE=0.239°C, RMSE=0.302°C).
• The GWO-XGBoost model achieved the highest accuracy (R²=0.540, RMSE=0.294°C, MAE=0.232°C) and computational efficiency (2.41s optimization time).
• SHAP analysis identified IRTave_TK, time zone (TZ), days in lactation (DOL), and body posture (BP) as critical predictors of CBT.

Conclusions:

• Machine learning models, particularly GWO-XGBoost, can accurately predict dairy cow CBT.
• Infrared thermography and physiological data are valuable inputs for heat stress management.
• This framework supports timely interventions to maintain dairy cow health and productivity.