# Application of multivariate techniques for estimating herd feed efficiency using chemical and near-infrared calibration models in dairy cattle

## TL;DR

This study compares two models for predicting feed efficiency in dairy cattle using chemical data and near-infrared spectroscopy, finding the NIR model more robust for practical use.

## Contribution

The study introduces a robust NIR spectroscopy-based model for estimating feed efficiency in dairy cattle, showing better generalizability than chemical composition models.

## Key findings

- The NIR model showed stable performance in external validation (R2 = 0.70) compared to the chemical model (R2 = 0.64).
- NIR spectroscopy is a fast and cost-effective method for preliminary nutritional assessment in dairy systems.
- Systematic bias in the chemical model suggests the need for further calibration improvements.

## Abstract

Summary: Near-infrared (NIR) spectroscopy is widely used in animal nutrition to evaluate the chemical composition of diets and feeds. In addition, its application is now also extensive and well proven for the evaluation of some indirect parameters, such as the digestibility of diets or animal efficiency. This study developed and compared 2 predictive models for estimating feed efficiency (FE) in dairy cattle using total mixed ration (TMR) data. The first model used the chemical composition of TMR, and the second applied NIR spectroscopy data analyzed with chemometric techniques. Both models showed good abilities to predict FE: The chemical model showed high accuracy in calibration but decreased performance in external validation, and the NIR model maintained stable predictive ability in validation. The greater robustness of the NIR model, as well as the fast and costeffective peculiarities of this approach, support its use for preliminary nutritional assessment in dairy systems. Icons made by authors from www.flaticon.com as follows: cow and table icons by Freepik; chemical analysis icon by imaginationlol; palette board icon by Prosymbols Premium; iMac icon by Those Icons; R icon by Becris; and cow icon by pbig.

Summary: Near-infrared (NIR) spectroscopy is widely used in animal nutrition to evaluate the chemical composition of diets and feeds. In addition, its application is now also extensive and well proven for the evaluation of some indirect parameters, such as the digestibility of diets or animal efficiency. This study developed and compared 2 predictive models for estimating feed efficiency (FE) in dairy cattle using total mixed ration (TMR) data. The first model used the chemical composition of TMR, and the second applied NIR spectroscopy data analyzed with chemometric techniques. Both models showed good abilities to predict FE: The chemical model showed high accuracy in calibration but decreased performance in external validation, and the NIR model maintained stable predictive ability in validation. The greater robustness of the NIR model, as well as the fast and costeffective peculiarities of this approach, support its use for preliminary nutritional assessment in dairy systems. Icons made by authors from www.flaticon.com as follows: cow and table icons by Freepik; chemical analysis icon by imaginationlol; palette board icon by Prosymbols Premium; iMac icon by Those Icons; R icon by Becris; and cow icon by pbig.

•Predictive NIR spectroscopic models were developed to forecast FE using only TMR data.•We confirmed NIR spectroscopy as a promising tool for rapid and cost-effective FE estimation.•The models show promise but require further refinement.

Predictive NIR spectroscopic models were developed to forecast FE using only TMR data.

We confirmed NIR spectroscopy as a promising tool for rapid and cost-effective FE estimation.

The models show promise but require further refinement.

Feed efficiency (FE) is an indicator of overall farm nutritional efficiency, helping farmers to identify any critical points in nutritional management. Particularly, FE is a measure of the ability of animals to convert feed into milk and it can be influenced by genetic, health, management, and nutritional factors. Higher FE allows results in reduced feed and maintenance costs and contributes to improved economic and environmental efficiency of dairy farms. This study aimed to develop and compare 2 predictive models for estimating FE in dairy cattle using data derived from the TMR: one based on its chemical composition and the other on near-infrared (NIR) spectral data. A total of 144 TMR samples were collected from farms in Po Valley from 2021 to 2024 and analyzed with an Fourier-transform NIR spectrometer. The spectral data were processed with chemometric techniques, including least absolute shrinkage and selection operator regression, in order to build a predictive model of FE. The model based on chemical composition showed strong calibration performance (R2 = 0.80, SE of cross-validation [SECV] = 0.13) but decreased in external validation (R2 = 0.64, SE of prediction [SEP] = 0.11), indicating the presence of systematic bias. Conversely, the NIR-based model maintained more stable performance between calibration (R2 = 0.73, SECV = 0.16) and external validation (R2 = 0.70, SEP = 0.09), with lower slope distortion and offset. The results suggest that although chemical data offer high accuracy in controlled conditions, the NIR model may be more robust and generalizable for practical, on-farm prediction of FE, offering potential decision support. However, further improvements in calibration are needed to reduce systematic errors and increase the accuracy of the model.