Fig. 2
From: Feature selection using distributions of orthogonal PLS regression vectors in spectral data
a Illustration of the empirical null distribution of realized \(\hat {\beta }_{OPLSb,i}\), which is close to normality, is shown. b Illustration of the empirical null distribution of realized \(\hat {\beta }_{OPLSb}\) for the variable with the largest confidence interval is shown. c Illustration of the regression vector \(\hat {\beta }_{OPLSb}\) is shown. The blue dots represent observed \(\hat {\beta }_{OPLSb,i}\), the green squares represent filtered-in variables according to the significance level αf, the red arrows with dotted lines represent achieved 1−αpm confidence intervals, and the black circles represent selected significant variables. d Zoomed-in version of (c) for the variables from 1 to 72 with 22 significant variables selected. The filtered-in variable 71 (x71), for example, was not selected since the observed \(\hat {\beta }_{OPLSb,i}\) is within the confidence interval of realized \(\hat {\beta }_{OPLSb,i}\)