Determining the digestibility of diets is crucial to develop efficient and ruminant sustainable production systems. Nevertheless, the methods currently available are not suitable for rapid and effective determination or estimation of digestibility. In recent years, models using near-infrared (NIR) spectroscopy have been developed to estimate digestibility directly from samples of feed or feces. Most of these models are global, meaning they rely on large, general datasets and may not always perform well for specific conditions or herds. In this study, several Local models are presented as an alternative to predicting the organic matter digestibility of the diet using a global model. These Local models are customized for each sample by selecting a subset of similar samples from a larger calibration database. The results show that several Local models can improve the accuracy of predictions compared with a global model. Among the models tested, the Locally weighted partial least squares regression model provided the best performance. This model is distinguished by its computational simplicity in comparison with the other models examined. The prediction error of the best Local model was slightly higher (+19%) compared to the measurement error obtained from the standard reference method.

Read the full article in the Journal of Animal Science.