Predicting nitrogen use efficiency, nitrogen loss and dry matter intake of individual dairy cows in late lactation by including mid-infrared spectra of milk samples

Table 5 The bias, slope, and random proportions of the mean square prediction error of best prediction models in with-herd validation for each trait¹

Trait	Algorithm	Model	MIR	MSPE²	RE	Bias%	Slope%	Random%
NUE	PLS	5	MSC	32.4(9.2)	0.21(0.03)	3.7(3.9)	4.2(6.6)	92.1(8.3)
	RR	2	MSC	35.6(9.4)	0.23(0.03)	2.3(2.0)	1.8(2.8)	95.9(3.2)
	SVM	5	SNV	16.1(2.3)	0.15(0.01)	5.7(5.1)	2.1(1.3)	92.2(5.7)
NL	PLS	1	SNV	1.1e-02(1.8e-03)	0.21(0.02)	4.9(5.0)	2.7(3.3)	92.5(3.7)
	RR	3	MSC	1.2e-02(1.2e-03)	0.21(0.01)	4.5(4.5)	6.5(6.4)	89.0(3.3)
	SVM	2	MSC	1.2e-02(5.7e-04)	0.22(0.004)	3.4(4.9)	6.4(4.4)	90.2(6.0)
DMI_a	PLS	3	MSC	18.7(3.5)	0.17(0.01)	4.8(4.1)	3.1(1.3)	92.1(4.7)
	RR	2	MSC	17.5(2.1)	0.17(0.01)	1.3(1.4)	1.6(1.6)	97.1(2.5)
	SVM	4	MSC	18.4(2.6)	0.17(0.01)	1.1(1.6)	6.1(4.3)	92.7(4.6)

¹NUE nitrogen use efficiency, NL nitrogen loss, DMI_a 3-d moving average of dry matter intake, PLS partial least squares, RR ridge regression, SVM support vector machine, MIR mid-infrared, MSC multiplicative scatter correction, SNV standard normal variate, MSPE mean square prediction error, RE relative error, Bias% proportion of error due to mean bias, Slope% proportion of error due to deviation of the slope from 1, Random% proportion of error explained by random error. Values between brackets indicate the standard deviation
²The unit of MSPE: % × % for NUE; kg × kg for NL and DMI_a

ISSN: 2049-1891