PreCowKetosis: A Shiny web application for predicting the risk of ketosis in dairy cows using prenatal indicators

Abstract

Ketosis is considered to be the most important metabolic disease affecting dairy herds, surpassing ruminal acidosis and milk fever. On dairy farms, assessment and monitoring of ketosis risk are necessary for the health management of dairy cows. Changes in the content of hematological and serum biochemical parameters are commonly used to monitor ketosis in dairy cows. However, the collection and detection of these indicators are complex and may lead to stress reactions in cows. This study attempted to predict the risk of ketosis in dairy cows using machine learning models based on noninvasive prenatal indicators of parity, body condition score, dystocia score, daily rumination time, daily activity, and season of calving. Results showed the extreme gradient boosting (XGBoost) model had the highest prediction ability and the most accuracy, compared to random forest (RF), support vector machine (SVM), artificial neural network (ANN), and K-nearest neighbor (KNN). In the XGBoost model, daily rumination time (60.15%) and daily activity (16.73%) were identified with the highest percentage contribution to the model, followed by parity (10.41%), body condition score (6.42%), season of calving (4.23%), and dystocia score (2.06%). The probability of ketosis increased with decreasing daily rumination time and daily activity. Moreover, parity 3+ and summer may also increase the probability of ketosis. Finally, an open Shiny web application for predicting the risk of ketosis in dairy cows based on the XGBoost model was developed (https://2xzl2o-neaop.shinyapps.io/PreCowKetosis/). The application PreCowKetosis can provide decision support for researchers and farmers to prevent ketosis in dairy cows.