Table 4.
Summary of predictive analysis of age as a binary response (0=young; 1=old) in Marmosets (training data) and Humans (blind test data) using the redox metabolites as predictor variables.
| MARS degree of interaction | Selected predictor variables | GCV | R2 | Train AUC | Test classification accuracy (%) | Test AUCb |
|---|---|---|---|---|---|---|
| 1 | h(0.0684-GSSG), | 0.217 | 0.227 | 0.678 | 59.5 | 0.383 |
| h(−0.217-CySS) | ||||||
| 2a | h(0.0684-GSSG) | 0.237 | 0.207 | 0.708 | 70.2 | 0.703 |
| h(−0.2169-CySS)b | ||||||
| h(1.081-GSH) | ||||||
| 3 | h(0.0684-GSSG), | 0.224 | 0.227 | 0.678 | 59.5 | 0.383 |
| h(−0.217-CySS) | ||||||
AUC=area under curve; GCV=generalized cross validation.
a
Best model based on AUC value from training stage.