Fig. 3

This figure illustrates feature selection using BCS for a simulated data set from the model in Fig. 2. a A biplot for the first orthogonal and predictive components is shown, clearing separating the two labels of response Y with a 95% correlation range for each label. b The corresponding loading plot, a zoomed-in version of the dotted-line rectangle in (a), is shown; the top 5% significantly contributing variables found from Layer 1, 2, and 3 are shown as red squares, blue diamonds, and black triangles, respectively. The eight strong variables (x₁ to x₈) in the plot are clearly distinguishable and positioned accordingly to the labels. For example, the position of x₁ is aligned in the direction of label 0, consistent with the behavior of x₁ in Fig. 2(b)