Table 1.
Misclassification rate (%) observed by technique and by significance filtering status (pre- vs. post-) throughout the 1000 baseline simulation studies and 1000 realistic simulation studies. The lowest median misclassification rate observed over each scenario type is shown in bold face.
| Method | Baseline Pre- | Baseline Post- | Realistic Pre- | Realistic Post- | ||||
|---|---|---|---|---|---|---|---|---|
| Mean ± SD | Median ± IQR | Mean ± SD | Median ± IQR | Mean ± SD | Median ± IQR | Mean ± SD | Median ± IQR | |
| PLS-DA | 15.0 ± 15.5 | 10.0 ± 28.3 | 13.1 ± 15.6 | 5.0 ± 25.0 | 32.2 ± 17.4 | 30.0 ± 26.7 | 25.0 ± 15.7 | 23.3 ± 24.6 |
| sPLS-DA | 13.1 ± 15.3 | 5.0 ± 25.0 | 12.1 ± 15.1 | 3.3 ± 23.3 | 19.7 ± 15.2 | 15.0 ± 23.3 | 22.0 ± 15.0 | 20.0 ± 21.7 |
| SVM | 23.2 ± 24.6 | 13.3 ± 41.7 | 10.4 ± 14.3 | 1.7 ± 19.6 | 22.8 ± 18.3 | 16.7 ± 26.7 | 13.3 ± 12.5 | 8.3 ± 13.3 |
| NNet | 22.0 ± 15.6 | 20.0 ± 25.0 | 15.9 ± 13.8 | 11.7 ± 21.7 | 29.9 ± 15.2 | 28.3 ± 21.7 | 23.3 ± 14.3 | 21.7 ± 21.7 |
| RF | 15.0 ± 14.8 | 10.0 ± 23.3 | 13.5 ± 14.4 | 7.5 ± 21.7 | 17.5 ± 16.0 | 11.7 ± 21.7 | 15.5 ± 15.0 | 10.0 ± 18.3 |
| k-NN | 20.2 ± 17.3 | 20.0 ± 28.3 | 21.9 ± 16.4 | 20.0 ± 25.0 | 41.3 ± 18.8 | 41.7 ± 26.7 | 41.6 ± 17.7 | 41.7 ± 26.7 |
| NB | 14.0 ± 15.3 | 8.3 ± 25.0 | 11.1 ± 14.6 | 1.8 ± 21.7 | 32.1 ± 18.2 | 30.0 ± 28.3 | 19.1 ± 14.8 | 15.0 ± 21.7 |
PLS-DA: Partial Least Squares-Discriminant Analysis; sPLS-DA: Sparse PLS-DA; SVM: Support Vector Machines; NNet: Artificial Neural Network; RF: Random Forest; k-NN: k-Nearest Neighbors; NB: Naïve Bayes