Machine Learning-Based Predictive Modeling for Early Detection of Subclinical Mastitis in Dairy Herds

Abstract: Subclinical mastitis (SCM) is a symptomless yet economically damaging udder infection that silently reduces milk yield and quality in dairy cattle, making early identification difficult through conventional visual inspection. This paper presents a machine-learning based predictive framework for early SCM detection using three routinely measurable and low-cost milk parameters: Somatic Cell Count (SCC), pH, and Electrical Conductivity (EC). A dataset of 5,000 milk samples (14.6% mastitis-positive) was enhanced through feature engineering, including logarithmic transformation and SCC-to-EC ratio, and rolling statistical features that capture infection trends over time. Logistic Regression and LightGBM classifiers were trained and compared, and the tuning of LightGBM using Optuna-based automated hyperparameter tuning achieved a test accuracy of approximately 98.8%, with precision of 0.93, recall of 0.96, and an F1-score of 0.94, outperforming Logistic baseline. The SHAP (Shapley Additive Explanations) analysis confirmed SCC and EC as the dominant predictive features, consistent with established veterinary knowledge, and validated the biological relevance of the engineered features. The resulting framework provides an accurate, interpretable, and cost- effective decision-support tool suitable for real-world dairy farm deployment.

Keywords: subclinical mastitis, machine learning, LightGBM, somatic cell count, electrical conductivity, explainable AI, SHAP, precision dairy farming