Harnessing ensemble models for improved lactation curve modeling: The R package EMOTIONS

Lactation curve modeling plays a key role in dairy production, supporting on-farm management decisions and the selection of animals with superior productivity and resilience. However, accurately choosing a single model to represent the diversity of lactation patterns across individuals remains challenging. In this context, we presented in this study a flexible ensemble modeling (EM) framework, implemented in the EMOTIONS R package, to improve the prediction of daily milk yield and related traits in dairy cows and ewes. EMOTIONS integrates a wide set of lactation curve models and allows users to generate ensembles by weighting model predictions based on multiple criteria. The package provides tools for model fitting, ensemble creation, visualization, milk yield imputation, detection of milk loss events, and the estimation of residual-based resilience indicators. The advantages of the ensembles were evident in subsets of animals with poor individual model fits, underscoring the value of EMs in capturing individual variation and reducing underfitting. In both ewes' and cows' data sets analyzed as examples, the EM advantages were clear for subsets of individual lactations with overall poor model fittings (based on BIC). Overall, EMs represent a robust and adaptable approach for modeling lactation data, offering improved predictive accuracy while retaining the ability to interpret biologically meaningful model parameters.