Table 4.
Holdout Dataset results: accuracies for binary and three-way diagnosis on the 171 participant Holdout Dataset using personal characteristic input data.
Diagnostic task | Input data | Classifier | Holdout accuracy (%) |
---|---|---|---|
Binary | Chance | 55.0 | |
PCs1 | Quadratic SVM | 69.0 | |
PCs2 | Linear SVM | 65.5 | |
Three-way | Chance | 55.0 | |
PCs1 | Logistic | 63.7 | |
PCs2 | Logistic | 59.1 |
Only the four best-performing Diagnostic Pipelines from Table 3 were tested on the Holdout Dataset. PCs1, personal characteristic data selection 1; PCs2, personal characteristic data selection 2. See section 2.3 for details. Holdout accuracy column shows accuracies derived from training each classifier on the 668 participant Training Dataset and then testing it on the Holdout Dataset. Chance baseline was derived by guessing control for every participant.