Table 3. Mean classification accuracies obtained by combining all data features together, after dimensionality reduction based on Principal Component Analysis (PCA Based combination), and after selecting only the 1% of variables with largest t values (t-thresholded combination).
Algorithm | PCA based combination | t-thresholded combination | |||||
---|---|---|---|---|---|---|---|
Mean accuracy | 5%limit | 95%limit | Mean accuracy | 5%limit | 95%limit | ||
Healthy vs. schizophrenia |
aridge | 0.6 | 0.516 | 0.673 | 0.733 | 0.686 | 0.776 |
blasso | 0.638 | 0.584 | 0.698 | 0.781 | 0.747 | 0.816 | |
celastic | 0.662 | 0.598 | 0.728 | 0.752 | 0.711 | 0.791 | |
dL0 norm | 0.713 | 0.655 | 0.771 | 0.761 | 0.717 | 0.798 | |
eSVC | 0.582 | 0.495 | 0.675 | 0.737 | 0.693 | 0.781 | |
fRDA | 0.658 | 0.593 | 0.725 | 0.792 | 0.758 | 0.823 | |
gGPC | 0.619 | 0.561 | 0.688 | 0.717 | 0.657 | 0.776 | |
hRF | 0.674 | 0.624 | 0.725 | 0.729 | 0.657 | 0.792 | |
Healthy vs. bipolar dis. |
ridge | 0.526 | 0.455 | 0.598 | 0.619 | 0.562 | 0.662 |
lasso | 0.553 | 0.5 | 0.612 | 0.596 | 0.561 | 0.639 | |
elastic | 0.549 | 0.484 | 0.626 | 0.619 | 0.586 | 0.646 | |
L0 norm | 0.557 | 0.495 | 0.62 | 0.564 | 0.518 | 0.618 | |
SVC | 0.518 | 0.461 | 0.574 | 0.608 | 0.583 | 0.632 | |
RDA | 0.529 | 0.464 | 0.603 | 0.608 | 0.563 | 0.663 | |
GPC | 0.522 | 0.459 | 0.585 | 0.623 | 0.561 | 0.677 | |
RF | 0.491 | 0.405 | 0.576 | 0.627 | 0.575 | 0.67 | |
Bipolar dis. vs. schizophrenia |
ridge | 0.492 | 0.446 | 0.537 | 0.641 | 0.601 | 0.678 |
lasso | 0.56 | 0.506 | 0.614 | 0.594 | 0.557 | 0.637 | |
elastic | 0.556 | 0.493 | 0.614 | 0.613 | 0.572 | 0.651 | |
L0 norm | 0.579 | 0.507 | 0.65 | 0.609 | 0.564 | 0.657 | |
SVC | 0.516 | 0.456 | 0.573 | 0.563 | 0.518 | 0.607 | |
RDA | 0.571 | 0.513 | 0.628 | 0.613 | 0.564 | 0.661 | |
GPC | 0.521 | 0.457 | 0.586 | 0.676 | 0.627 | 0.718 | |
RF | 0.529 | 0.453 | 0.599 | 0.637 | 0.583 | 0.686 |
Limits for 95% confidence intervals are based on bootstrap.
aridge: Ridge regression,
blasso: Lasso regression,
celastic: Elastic net regularization,
dL0-norm: L0-norm regularization,
eSVC: Support vector classifier,
fRDA: Regularized discriminant analysis,
gGPC: Gaussian process classifier,
hRF: Random forest.