Genomic prediction of unordered categorical traits: an application to subpopulation assignment in German Warmblood horses

Table 3 Prediction accuracies and scaled Brier scores for subpopulation assignment

Base	Contrast	Accuracy					sBS
Base	Contrast	1	2	3	4	Total	1	2	3	4	Total
Ridge	Regression
1234	One-vs-one	0.97	0.99	0.00	0.86	0.74	0.86	0.97	−0.43	0.66	0.58
1234	One-vs-all	0.86	1.00	0.54	0.86	0.84	0.69	0.98	0.18	0.68	0.68
234	One-vs-one		1.00	0.87	0.75	0.93		0.97	0.72	0.59	0.85
234	One-vs-all		1.00	0.87	0.93	0.95		0.98	0.70	0.77	0.87
134	One-vs-one	0.85		0.66	0.64	0.76	0.66		0.31	0.36	0.50
134	One-vs-all	0.85		0.67	0.82	0.78	0.66		0.32	0.59	0.52
124	One-vs-one	0.97	1.00		0.66	0.96	0.93	0.99		0.38	0.92
124	One-vs-all	0.97	1.00		0.80	0.97	0.92	0.99		0.61	0.94
123	One-vs-one	0.84	1.00	0.54		0.83	0.65	0.96	0.09		0.63
123	One-vs-all	0.85	1.00	0.53		0.83	0.65	0.97	0.07		0.63
SVM
1234	One-vs-one	0.86	1.00	0.50	0.59	0.81
234	One-vs-one		1.00	0.85	0.73	0.93
134	One-vs-one	0.87		0.62	0.59	0.75
124	One-vs-one	0.96	1.00		0.59	0.96
123	One-vs-one	0.86	1.00	0.50		0.82

The proportion of correct classifications to all classifications (Accuracy) and the scaled Brier Scores (sBS) per class and over all samples (Total) for the classification of subpopulation assignment using leave-one-out cross-validation. Base indicates the classes that were included in training and testing. Numeric indicators for subpopulations are: 1 Hanoverian, 2 Holsteiner, 3 Oldenburger, 4 Trakehner

ISSN: 1297-9686