Fig. 1From: Feature selection using distributions of orthogonal PLS regression vectors in spectral dataa A data model to generate 1000 variables has three layers of variable interation that contribute to the response Y. Layer 1 contains 30 variables, including eight strong variables (x1 to x9) and the other group-wise variables (x10 to x30). Each of the variables in Layer 1 is determined by three variables in Layer 2, and each variable in Layer 2 by three variables in Layer 3. The remaining 610 variables are randomly and independently assigned. b Two strong variables x1 and x5 separate the two labels of Y. c Three variables in Layer 1 jointly separate the two labels of response YBack to article page