Table 3. Performance of SVM models on NIPT prediction using different parameter setting.
Chr21 | Group "N" & "P" | Group "Unclassified" | ||||||||
Model e | Real status | Support vector number | Prediction | Sens. f | Spec. | Prediction | Sens. | Spec. | ||
N | P | N | P | |||||||
SVM-RBF-opt | N | 365 | 4672 | 0 | 100.00% | 100.00% | 57 | 0 | 100.00% | 100.00% |
P | 19 | 0 | 19 | 0 | 4 | |||||
SVM-linear-opt | N | 2 | 4672 | 0 | 100.00% | 100.00% | 57 | 0 | 0.00% | 100.00% |
P | 2 | 0 | 19 | 4 | 0 | |||||
SVM-RBF-opt-w | N | 478 | 4672 | 0 | 100.00% | 100.00% | 57 | 0 | 100.00% | 100.00% |
P | 19 | 0 | 19 | 0 | 4 | |||||
SVM-linear-opt-w | N | 2 | 4672 | 0 | 100.00% | 100.00% | 57 | 0 | 0.00% | 100.00% |
P | 2 | 0 | 19 | 4 | 0 | |||||
Chr18 | Group "N" & "P" | Group "Unclassified" | ||||||||
Model | Real status | Support vector number | Prediction | Sens. | Spec. | Prediction | Sens. | Spec. | ||
N | P | N | P | |||||||
SVM-RBF-opt | N | 106 | 4697 | 0 | 100.00% | 100.00% | 44 | 0 | 100.00% | 100.00% |
P | 7 | 0 | 7 | 0 | 4 | |||||
SVM-linear-opt | N | 2 | 4697 | 0 | 85.71% | 100.00% | 44 | 0 | 0.00% | 100.00% |
P | 2 | 1 | 6 | 4 | 0 | |||||
SVM-RBF-opt-w | N | 303 | 4697 | 0 | 100.00% | 100.00% | 44 | 0 | 100.00% | 100.00% |
P | 6 | 0 | 7 | 0 | 4 | |||||
SVM-linear-opt-w | N | 3 | 4697 | 0 | 85.71.00% | 100.00% | 44 | 0 | 0.00% | 100.00% |
P | 1 | 1 | 6 | 4 | 0 | |||||
Chr13 | Group "N" & "P" | Group "Unclassified" | ||||||||
Model | Real status | Support vector number | Prediction | Sens. | Spec. | Prediction | Sens. | Spec. | ||
N | P | N | P | |||||||
SVM-RBF-opt | N | 1976 | 4706 | 0 | 100.00% | 100.00% | 42 | 0 | NA | 100.00% |
P | 4 | 0 | 4 | 0 | 0 | |||||
SVM-linear-opt | N | 2 | 4706 | 0 | 100.00% | 100.00% | 42 | 0 | NA | 100.00% |
P | 2 | 0 | 4 | 0 | 0 | |||||
SVM-RBF-opt-w | N | 2070 | 4706 | 0 | 100.00% | 100.00% | 42 | 0 | NA | 100.00% |
P | 4 | 0 | 4 | 0 | 0 | |||||
SVM-linear-opt-w | N | 2 | 4706 | 0 | 100.00% | 100.00% | 42 | 0 | NA | 100.00% |
P | 2 | 0 | 4 | 0 | 0 |
Four types of SVM models were compared in both internal and external validation for each of chromosome 13/18/21.
e w means employing class weight to adjust parameter C; opt means employing optimization for parameters C and gamma in cross validation.
f Sens. is short for sensitivity; Spec. is short for specificity.