International Journal of Scientific Research and Engineering Development

Subclinical mastitis continues to be one of the key silent risks to dairy farm efficiency, developing symptomlessly and causing substantial losses in milk production and its quality. In this paper, we introduce a practical implementation of the developed machine learning-based approach for the timely diagnosis of subclinical mastitis through the analysis of the physicochemical characteristics of milk: Somatic Cell Count (SCC), pH, and Electrical Conductivity (EC). The system incorporates the elements of feature engineering (log-SCC, SCC/EC ratio, and average SCC), hyperparameter optimization based on the Optuna package, and SHAP explainability for better understanding of the decision process. Two approaches, Logistic Regression and hyperparameter optimized LightGBM model, were tested and evaluated using the real-life dataset containing information about 5000 milk samples. The optimized LightGBM model reached the highest accuracy of 98.3% and F1-score equal to 0.94 and showed the best performance compared to baseline models. The obtained results were further supported by SHAP analysis proving SCC and EC as the key factors influencing the probability of infection.

Gouramma Naganuri, Shweta Adiveppa Budihal, Spoorti Arun Naik, Srushti S N, Mrs. Deepika A B, "Smart Milk Analysis for Early Prediction of Subclinical Mastitis in Dairy Herds" International Journal of Scientific Research and Engineering Development, V9(4): Page(997-1003) July-August 2026. ISSN: 2581-7175. www.ijsred.com. Published by Scientific and Academic Research Publishing.