Predictive Models for Weekly Cattle Mortality after Arrival at a Feeding Location Using Records, Weather, and Transport Data at Time of Purchase

Abstract

Feedlot mortality negatively affects animal welfare and profitability. To the best of our knowledge, there are no publications on predictive models for weekly all-cause mortality in feedlot cattle. In this study, random forest models to predict weekly mortality for cattle purchase groups (n = 14,217 purchase groups; 860,545 animals) from arrival at the feeding location (Day 1) to Day 42 and cumulative mortality from Day 43 until slaughter were built using records, weather, and transport data available at the time of purchase. Models were evaluated by calculating the root mean squared error (RMSE) and accuracy (as defined as the percent of purchase groups that had predictions within 0.25% and 0.50% of actual mortality). The models had high accuracy (>90%), but the RMSE estimates were high (range = 1.0% to 4.1%). The best predictors were maximum temperature and purchase weight, although this varied by week. The models performed well among purchase groups with low weekly mortality but performed poorly in high mortality purchase groups. Although high mortality purchase groups were not accurately predicted utilizing the models in this study, the models may potentially have utility as a screening tool for very low mortality purchase groups after arrival. Future studies should consider building iterative models that utilize the strongest predictors identified in this study.

Keywords: beef cattle; feedlot health; mortality; predictive modeling; random forests.