Skip to main content

Table 3 Accuracies and unbiasedness of genomic prediction on TNB and NBA from seven methods in 20 replicates of 5-fold CV

From: Using machine learning to improve the accuracy of genomic prediction of reproduction traits in pigs

Hyper-parameters

Method

TNB1

NBA2

Accuracy3

Unbiasedness4

Accuracy3

Unbiasedness4

 

GBLUP

0.248a ± 0.026

0.958 ± 0.132

0.208a ± 0.025

0.931 ± 0.142

 

ssGBLUP

0.251a ± 0.026

0.901 ± 0.121

0.221ab ± 0.026

0.844 ± 0.113

 

BayesHE

0.243a ± 0.025

1.015 ± 0.148

0.207a ± 0.026

1.009 ± 0.171

Tuning

SVR

0.295b ± 0.025

1.23 ± 0.119

0.254b ± 0.023

1.106 ± 0.11

KRR

0.295b ± 0.025

1.266 ± 0.125

0.256b ± 0.023

1.151 ± 0.113

RF

0.270ab ± 0.029

1.229 ± 0.152

0.248ab ± 0.028

1.188 ± 0.147

Adaboost.R2_SVR

0.293b ± 0.025

1.363 ± 0.138

0.254b ± 0.024

1.256 ± 0.131

Adaboost.R2_KRR

0.292b ± 0.025

1.344 ± 0.136

0.258b ± 0.024

1.249 ± 0.129

Default

SVR

0.255 ± 0.027

1.275 ± 0.147

0.224 ± 0.023

1.098 ± 0.126

KRR

0.264 ± 0.025

1.007 ± 0.108

0.222 ± 0.024

0.879 ± 0.101

RF

0.246 ± 0.028

1.064 ± 0.142

0.225 ± 0.027

1.002 ± 0.128

Adaboost.R2_SVR

0.273 ± 0.024

0.998 ± 0.106

0.228 ± 0.026

0.822 ± 0.099

Adaboost.R2_KRR

0.254 ± 0.024

0.759 ± 0.085

0.209 ± 0.027

0.636 ± 0.085

  1. 1 TNB: total number of piglets born
  2. 2 NBA: number of piglets born alive
  3. 3 Accuracy: the correlation between corrected phenotypes and predicted values of the validation population;
  4. 4 Unbiasedness: the regression of corrected phenotypes onto the predicted values
  5. The different superscript of accuracy indicates the significant difference by the Hotelling-Williams test