Comparison between LASSO and RT methods for prediction of generic E. coli concentration in pastured poultry farms

• Published in: Food research international Vol. 161; p. 111860
• Main Authors: Xu, Xinran, Rothrock Jr, Michael J., Reeves, Jaxk, Dev Kumar, Govindaraj, Mishra, Abhinav
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01.11.2022
• Subjects: E. coli concentration; Farm management practices; LASSO; Pastured poultry farms; Predictive modeling; E. coli concentration; RT; Predictive modeling; Farm management practices; Pastured poultry farms; LASSO
• ISSN: 0963-9969; 1873-7145
• DOI: 10.1016/j.foodres.2022.111860

[Display omitted] •LASSO and CART models are useful to predict E. coli prevalence.•The animal source was an important factor affecting E. coli prevalence.•The type of chemical used to rinse chicken carcasses affected E. coli levels.•Seasonal effect of E. coli population was observed in feces samples. Though most strains of E. coli are non-pathogenic components of the intestinal microbiome, certain pathogenic E. coli strains are the cause of diseases and outbreaks. Poultry is identified as a common reservoir for pathogenic E. coli. It is important to identify farm practice factors associated with E. coli in the pastured poultry environment. The objective of this study is to develop models that can predict E. coli levels and to select farm practice factors contributing to E. coli concentration in pastured poultry farms. Five kinds of samples: feces, soil, whole carcass rinse after processing (WCR-P), final product after chilling and storage (WCR-F), and ceca samples were collected for E. coli counts from 11 pastured poultry farms in the southeastern US. The regression tree (RT) and least absolute shrinkage and selection operator (LASSO) methods were applied to data from each sample type. The farm management practices and processing factors such as source of eggs, type of feed used, appearance of other animals on farm, chilling method, and storage time and temperature were all considered as possible explanatory factors in the models. Models were developed to predict the levels of E. coli and to select the most important factors used in predicting E. coli population. Model performances were compared using root mean square error (RMSE). For feces samples, average number of birds and animal source were the two most important variables affecting E. coli population by LASSO. The RT selected animal source, brood feed, day of year, flock age in days, and flock size as the most important variables in predicting E. coli concentration. The RMSE (in log10 scale) under LASSO was 0.974, while under RT it was 1.032 for feces samples. The predictive models provide practical and effective tools to predict E. coli population and to identify farm practice factors that affect E. coli levels.