Abstract Background Loading and transporting pigs from farms to abattoirs is the final stage before slaughter and marketing. Transit losses of market-weight pigs refer to pigs that die or become non-ambulatory during transportation, resulting in both economic losses and animal welfare concerns. Measuring the causes of transport loss is complex due to the multifactorial factors influencing it, while data on those factors are usually disconnected. Objective To jointly evaluate site- and truck-level factors associated with transport losses per load and to develop a machine-learning model to forecast load-level loss risk. Methods We analyzed 30,291 transport records from 1,928 finishing batches across 348 sites in the US Midwest (January 2022–May 2025). R scripts integrated finishing closeouts, transport summaries, and loss outcomes. A negative binomial mixed-effects model was fitted with the number of transit losses per load as the response, and fixed effects were selected through manual stepwise procedures, retaining variables significant at p 0.05. Estimated marginal means were compared across categories for the variables included in the final model. Thereafter, an XGBoost model approach using the original data and an 80:20 cross-validation split was applied to predict loads with high or low transit losses. A load-level transit loss risk score was created by scaling the negative predictive value from 0 to 10. Results The average transit loss percentage was 1.09% (95% CI: 1.08–1.11) per load. Of 36 initial variables, 16 remained in the final model. Sites equipped with wet–dry feeder combinations showed higher transit loss percentages. At the batch level, batches with pigs of lower body weights upon entering the site, and higher average daily feed intake (ADFI) were more likely to experience transport losses. For transport factors, loads had a shorter traveling distance, and transported during fall and winter, or closer to the group-close date, also exhibited higher transit loss risk. At the truck level, a heavier overall load weight, a greater proportion of heavy pigs, and a more unbalanced weight distribution (very light or very heavy pigs) were associated with higher probabilities of losses during transport. The XGBoost model achieved 0.7005 accuracy (95% CI: 0.6888–0.712), with 0.6868 sensitivity and 0.7139 specificity. The XGBoost-based 0–10 score reflected the likelihood of low transport losses, where higher scores denoted lower risk. Loads scoring above 7.5 were mainly associated with low losses, while those below 2.5 corresponded to higher losses. Conclusion Pigs under higher health risk or experiencing lower growth performance exhibit a greater risk of transit loss. Cold seasons are found to be riskier in this study. Adjusting transport timing and balancing load composition can reduce losses and improve animal welfare. Regarding machine learning, it offers a probability-based pre-assessment of transit losses for truckloads by assigning a load-risk score.
Huang et al. (Wed,) studied this question.