machine learning algorithms

In dairy industry, high producing fresh dairy cows commonly experience adipose tissue mobilization to support their energy requirements. Precise prediction of blood beta-hydroxybutyric acid (BHBA) concentration could significantly enhance the cow health and welfare, therefore, this study aimed to identify the key factors influencing BHBA levels and develop predictive models based on nutritional and performance data in fresh dairy cows. In this trial, four years data from 325 fresh Holstein cows were collected and analyzed. Various machine learning algorithms, including decision trees, random forests, Lasso and ridge regression models, as well as boosting and bagging techniques, were employed to estimate BHBA levels and identify the influential factors. These algorithms were assessed using the coefficient of determination (R²). The random forest model demonstrated the lowest error, with a mean absolute error of 0.02, while the linear model exhibited the highest error, with a mean absolute error of 1.25. It was found that factors including milk production, previous lactation days in milk (DIM), sampling day, body weight change, BCS at parturition, and the amount and type of dietary fat, as well as overall diet quality were highly significant for estimating blood BHBA levels (P<0.05). Notably, the results indicated that cows with a BCS of 3 or lower, as well as those with a score of 3.75, are crucial categories for predicting BHBA. Additionally, the level and type of fatty acids in the diet, particularly lauric (C12:0), palmitic (C16:0), linolenic (C18:3), and oleic acids (C18:1), had significant influence on BHBA in fresh cows (P<0.05). These findings highlight the importance of integrating these critical factors into predictive models to enhance metabolic health monitoring and improve dairy herd management practices.