Research on the prediction model of mastitis in dairy cows based on time series characteristics

Rui Guo Dai Yongqiang ^* Junjie Hu

• Gansu Agricultural University, Lanzhou, China

Mastitis is an important disease affecting the global dairy industry, resulting in significant losses to dairies due to reduced milk quality and production. With increasing consumer interest in food safety and the a ppropriate use of antibiotics, early identification of cows at risk for the disease has become an urgent issue. In particular, subclinical mastitis is more difficult to detect due to the lack of visible symptoms, and its early warn ing is particularly important. In this study, a time series prediction method was used to predict the risk of masti tis in dairy cows based on machine learning techniques. The study data were obtained from 4000 cows from la rge farms in the Hexi region of Gansu, and the time series features were constructed to predict the health status of each cow in June by utilizing the production indicators such as milk yield, fat percentage and protein perce ntage in two consecutive months in April and May. To fully utilize the time-series features, we constructed a m ultivariate enlistment including raw indicator values, monthly rates of change, and statistical features. After dat a preprocessing and balancing, data from 2821 cows were selected for model training. By comparing the perfo rmance of several prediction models, including XGBoost, gradient boosting tree, support vector machine, K-ne arest neighbor ,logistic regression and Long Short-Term Memory, it was found that the XGBoost model perfor med the best, with an area under the ROC curve of 0.75 and an accuracy rate of 71.36%. Characteristic import ance analysis showed that May milk yield (22.29%), standard deviation of fat percentage (20.27%) and rate of change of fat percentage (19.87%) were the key time-series indicators affecting the prediction results.The SHa pley Additive exPlanations (SHAP)value analysis further confirmed the predictive value of these time-series c haracteristics, which provided a clear monitoring focus for farm managers. This study provides an effective ear ly warning tool for disease prevention in dairy farms through time series modeling, which has important practi cal application value.