| 5-fold cross-validation |  | Independent test set |
---|
Classifier
|
Features
|
Accuracy
|
Accuracy
|
LDA | 19 | 79.5% | 78.7% |
SVM-linear | 25 | 80.1% | 78.9% |
SVM-RBF | 7 | 81.5% | 80.9% |
- LDA, SVM (linear kernel), and SVM (radial basis kernel) classifiers were first assessed by 5-fold cross validation accuracy for a range of model sizes. The minimum number of features that gave a cross-validation accuracy within one SEM of the maximum accuracy for all model sizes was determined, and is shown in the first column of the table, along with the corresponding cross-validation accuracy. Once the optimal parsimonious model size (number of features) was determined from cross-validation, the classifiers were applied to the independent test set; the test-set accuracy is shown in the right-most column of the table.