Table 1 Predictive performance of different modeling approaches for manure N output using stepwise and variance inflating factors (VIF) as feature selection methods.

From: Can machine learning algorithms perform better than multiple linear regression in predicting nitrogen excretion from lactating dairy cows

Models1

RMSE2

CCC3

Stepwise4

VIF5

Stepwise4

VIF5

MLR

44.7b

/

0.60ab

/

RFR

46.8b

38.3b

0.58a

0.68b

SVR

44.9b

45.3c

0.64b

0.63a

ANN

34.7a

28.5a

0.70c

0.78c

Sig.6

P < 0.01

P < 0.01

P < 0.01

P < 0.01

  1. a,b,cMeans within a column with different superscripts differ (P < 0.05).
  2. 1MLR multiple linear regression; RFR random forests regression; SVR support vector regression; ANN artificial neural network.
  3. 2RMSE root mean square error (obtained by tenfold cross validation).
  4. 3CCC concordance correlation coefficients (obtained by tenfold cross validation).
  5. 4The features selected by using stepwise methods were NI (N intake), LW (live weight) and MY (milk yield).
  6. 5The features selected by using variance inflating factors (VIF) method were NI (N intake), LW (live weight), MY (milk yield), FP (forage proportion), DNC (diet N concentration) and DMEC (diet metabolizable energy concentration).
  7. 6The significance was determined by one-way analysis of variance and followed by Tukey’s Honest Significant Difference (HSD) test (n = 10, α = 0.05).
Back to article page