Machine learning for ranking multivariate variables in cattle breeds raised in Paraguayan wetlands

ABSTRACT This study focuses on the performance of cows for meat production raised in the wetlands of Paraguay, examining five cattle genotypes: Brahman, Brangus, and Nelore, as well as two local breeds at risk of extinction. The main objective is to identify and rank phenotypic variables, including blood, clinical, hair, and health variables, demonstrating causal linkage with the live weight of the cows analyzed. Initially, high correlations were identified between different variables included in this study; then, using advanced Machine learning (ML) techniques and the application of Shapley additive explanations (SHAP), a deeper understanding was provided of the factors strongly associated with adaptability in these environments, and, therefore, the respective zootechnical performance. The association between cattle genotypic components linked with the season of the year proved to be the most influential factor on cattle live weight. Variables such as hair length, hematocrit, phosphatase, phosphorus, creatine phosphokinase, creatinine, protein, cortisol, calcium, and the presence of endoparasites were highlighted, demonstrating their hierarchical importance for animal selection. ML models are effective tools for establishing hierarchies of relevance in complex phenotypic multivariable, which is crucial in breeding programs for different zootechnical species and in special and specific environments like wetlands.