Table 3.
Training Dataset results: accuracies for binary and three-way diagnosis using various classifiers with personal characteristic data from the 668 participant Training Dataset.
| Diagnostic task | Input data | Classifier | Accuracy (%) | P value |
|---|---|---|---|---|
| Binary | Chance | 64.2 ± 3.8 | ||
| Personal characteristic | Logistic | 73.7 ± 5.1 | 1 × 10−5 | |
| data selection 1 (PCs1) | Linear SVM | 74.4 ± 4.6 | 9 × 10−6 | |
| Quadratic SVM | 74.5 ± 6.2 | 4 × 10−5 | ||
| Cubic SVM | 74.4 ± 4.8 | 2 × 10−6 | ||
| RBF SVM | 64.2 ± 3.8 | NA | ||
| Personal characteristic | Logistic | 74.0 ± 5.0 | 1 × 10−5 | |
| data selection 2 (PCs2) | Linear SVM | 75.0 ± 4.5 | 7 × 10−6 | |
| Quadratic SVM | 74.2 ± 6.3 | 5 × 10−5 | ||
| Cubic SVM | 73.8 ± 5.0 | 3 × 10−5 | ||
| RBF SVM | 64.2 ± 3.8 | NA | ||
| Three-way | Chance | 64.2 ± 3.8 | ||
| Personal characteristic | Logistic | 68.7 ± 8.1 | 0.020 | |
| data selection 1 (PCs1) | Linear SVM | 66.9 ± 7.1 | 0.038 | |
| Quadratic SVM | 68.6 ± 7.5 | 0.006 | ||
| Cubic SVM | 68.6 ± 8.2 | 0.020 | ||
| RBF SVM | 64.2 ± 3.8 | NA | ||
| Personal characteristic | Logistic | 69.0 ± 8.3 | 0.020 | |
| data selection 2 (PCs2) | Linear SVM | 66.9 ± 7.2 | 0.050 | |
| Quadratic SVM | 68.9 ± 7.9 | 0.009 | ||
| Cubic SVM | 67.5 ± 7.7 | 0.046 | ||
| RBF SVM | 64.2 ± 3.8 | NA |
PCs1, personal characteristic data selection 1; PCs2, personal characteristic data selection 2. See section 2.3 for details. Accuracy column shows mean accuracies ± standard deviations across 10-fold cross validation. Best result in a given category is shown in bold font. P value column shows p values for one-tailed, paired samples t-tests (df = 9) of accuracies against chance baseline (guessing control for every participant). NA (not available) indicates that the t-test was undefined in the case of SVM (RBF), which always guessed control and produced predictions identical to chance baseline.