Comparative Assessment of Lactation Models for Evaluating Herd Productivity: Classical model and Machine Learning techniques

Lactation models are one of the pivotal tools in the evaluation of a herd's productive capacity and the individual productivity of cows. In this study, we propose a comparative assessment of the performance of various lactation models. Wood and Milkbot are recognized lactation curve models and are based on exponential equations. Newer models include models based on artificial intelligence belonging to the autoencoder class. A comparison was carried out on a dataset, composed of 31,693,777 observations and 3498 Holstein cows coming from 91 farms. Different metrics were used to evaluate the model's performance, such as the coefficient of determination (R²), the percentage of milk error and the RMSE. The results reported a better performance of Wood between the two exponential equations for all metrics. However, The autoencoder with 30 neurons performed better than the other models in terms of R², and RMSE. These results highlight the ability to reconstruct milk yields from the derived low-dimensional representations, outperforming earlier developed models. Through this investigation, we aimed to provide valuable insights into the effectiveness of novel prediction models and shed light on their applicability in assessing lactation dynamics.