Abstract
Combining milk Mid-infrared (MIR) spectra with other cow descriptive and performance variables improves the prediction accuracy of total dry matter intake (TDMI) in dairy cows above using the milk MIR spectra only. However, the improvement in accuracy of prediction depends also on data fusion and prediction methods. Our objective was to evaluate low- and high-level data fusion strategies with partial least square (PLS) and artificial neural networks (ANN). Prediction equations were developed using a total of 384 observations and 4 different models. Predictions were performed using ANN, PLS and multiple linear regression (MLR). MLR was implemented using the variables A (milk yield, concentrate dry matter intake, week of lactation, parity and metabolic body weight). ANN was implemented using variables A, using only milk MIR spectra data or combining milk MIR spectra and variables A. PLS was implemented using variables A or using the milk MIR spectra and variables A. The high-level data fusion method using ANN had the highest out of sample coefficient of determination (R2) of 0.62. Direct merging of milk MIR spectra data with variables A explained better the TDMI than the milk MIR spectra data using both ANN and PLS prediction methods. Generally, the ANN methods perform better than PLS at low as well as at high level data fusion. Merging both milk MIR spectra data and animal variables at high level fusion enables to exploit better the information in the data. This method has thus a great potential to predict TDMI in commercial dairy farms. However, our finding needs additional evidence using large dataset and external validation.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have