Skip to main content

Table 3 Performance of each method of OPLSR a, OPLSR b, FDR, and Lasso is shown with respect to each parameter (q-value for FDR, λ for Lasso, and αf for the other two). The precision and the number of selected variables were denoted by P and N, respectively. The boldfaced numbers indicate the ones which outperformed the others

From: Feature selection using distributions of orthogonal PLS regression vectors in spectral data

Method MSE Q2 P N
OPLSRa αf=0.01 12,332 0.86 0.871 11.0
  αf=0.05 8,989 0.89 0.701 16.2
  αf=0.10 6,829 0.91 0.640 22.7
OPLSRb αf=0.01 10,896 0.88 0.915 11.7
  αf=0.05 8,704 0.91 0.860 28.3
  αf=0.10 7,081 0.92 0.818 41.2
FDR q=0.01 33,038 0.64 0.750 9.0
  q=0.05 12,123 0.86 0.714 36.6
  q=0.10 11,194 0.85 0.536 40.1
Lasso λ=0.01 7,120 0.91 0.545 6.6
  λ=0.10 10,796 0.89 0.477 4.0
  λ=0.50 14,028 0.81 0.496 2.2
\