Random Forest Modeling To Identify Key Farm-To-Fork Factors Influencing Campylobacter Ecology In Pastured Poultry Systems

Authors: KIM, MINHO - Orise Fellow; AL HAKEEM, WALID - University Of Georgia; Rothrock Jr, Michael

Submitted to: Poultry Science
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 6/9/2026
Publication Date: 6/10/2026
Citation: Kim, M., Al Hakeem, W.G., Rothrock Jr, M.J. 2026. Random Forest Modeling To Identify Key Farm-To-Fork Factors Influencing Campylobacter Ecology In Pastured Poultry Systems. Poultry Science. https://doi.org/10.1016/j.psj.2026.107274.
DOI: https://doi.org/10.1016/j.psj.2026.107274

Interpretive Summary: Campylobacter in poultry flocks poses significant food safety challenges, yet the drivers of its prevalence and load across the farm-to-fork continuum are not well understood. This study applied two-part random forest models to identify key factors influencing Campylobacter prevalence and load in pastured poultry systems. Data were collected from 11 farms in the southeastern United States between 2014 and 2017. The study included 1,942 broiler samples across five types: pasture soil, feces, ceca, whole carcass rinse after processing (WCR-P), and after storage (WCR-F). Two predictor sets were evaluated by 5-fold stratified cross-validation: farming practices with soil physicochemical properties and meteorological variables. Partial dependence plots were used to assess directional trends for key predictors. Models showed strong overall classification and regression performance across most sample types. Subsequent analyses focused on feces, ceca, and WCR-F. Soil was excluded due to low sensitivity and WCR-P was excluded due to intermediate biological position in the production, respectively. Classification models consistently identified farm as the dominant predictor of Campylobacter prevalence. This can reflect the cumulative effect of site-specific management and processing practices unique to each farm. Flock age was the second most important predictor for fecal samples. Prevalence increased as birds matured. Day of year was another leading predictor for cecal and WCR-F samples, predicting the highest prevalence during summer. Regression models identified flock age as the top predictor of Campylobacter load in feces and the second most important predictor in ceca. Campylobacter load declined with bird age in feces whereas cecal loads continued to accumulate as an internal reservoir. Meteorological models showed that sustained high wind speed reduced fecal Campylobacter prevalence. Lower rolling-average humidity and higher rainfall on the sampling day were each associated with lower fecal loads. These findings indicate that Campylobacter prevalence and load are affected by distinct drivers at each production stage. Targeted interventions such as optimizing flock management schedules and implementing farm-specific biosecurity measures could improve Campylobacter control throughout the pastured poultry production continuum.

Technical Abstract: Campylobacter in poultry flocks poses significant food safety challenges, yet the drivers of its prevalence and load across the farm-to-fork continuum are not well understood. This study applied two-part random forest models to identify key factors influencing Campylobacter prevalence and load in pastured poultry systems. Data were collected from 11 farms in the southeastern United States between 2014 and 2017. The study included 1,942 broiler samples across five types: pasture soil, feces, ceca, whole carcass rinse after processing (WCR-P), and after storage (WCR-F). Two predictor sets were evaluated by 5-fold stratified cross-validation: farming practices with soil physicochemical properties and meteorological variables. Partial dependence plots were used to assess directional trends for key predictors. Models showed strong overall classification and regression performance across most sample types. Subsequent analyses focused on feces, ceca, and WCR-F. Soil was excluded due to low sensitivity and WCR-P was excluded due to intermediate biological position in the production, respectively. Classification models consistently identified farm as the dominant predictor of Campylobacter prevalence. This can reflect the cumulative effect of site-specific management and processing practices unique to each farm. Flock age was the second most important predictor for fecal samples. Prevalence increased as birds matured. Day of year was another leading predictor for cecal and WCR-F samples, predicting the highest prevalence during summer. Regression models identified flock age as the top predictor of Campylobacter load in feces and the second most important predictor in ceca. Campylobacter load declined with bird age in feces whereas cecal loads continued to accumulate as an internal reservoir. Meteorological models showed that sustained high wind speed reduced fecal Campylobacter prevalence. Lower rolling-average humidity and higher rainfall on the sampling day were each associated with lower fecal loads. These findings indicate that Campylobacter prevalence and load are affected by distinct drivers at each production stage. Targeted interventions such as optimizing flock management schedules and implementing farm-specific biosecurity measures could improve Campylobacter control throughout the pastured poultry production continuum.