Table 2 The average numbers of selected variables in the layers for each method of OPLSR, according to \(\hat {\beta }_{OPLSa}\) (denoted by OPLSR a) and \(\hat {\beta }_{OPLSb}\) (denoted by OPLSR b), and FDR; and significance level α, meaning αf for OPLSR b and q-value for FDR, are shown. This shows that OPLSR a found more variables consistently while OPLSR b worked comparably to FDR
From: Feature selection using distributions of orthogonal PLS regression vectors in spectral data
α |  | 0.01 |  | 0.05 |  | 0.10 | ||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Â | Â | Layer | Â | Layer | Â | Layer | ||||||
| Â | Â | 1 | 2 | 3 | Â | 1 | 2 | 3 | Â | 1 | 2 | 3 |
OPLSR a | Â | 24.4 | 0.837 | 0.580 | Â | 25.0 | 3.79 | 5.61 | Â | 25.1 | 5.32 | 10.0 |
OPLSR b | Â | 18.1 | 0.190 | 0.137 | Â | 20.8 | 0.987 | 1.24 | Â | 22.9 | 2.85 | 3.68 |
FDR | Â | 21.3 | 0.020 | 0.000 | Â | 22.1 | 0.100 | 0.027 | Â | 22.8 | 0.103 | 0.010 |