Summary statistics for the absolute values of validation set element prediction error (VSEPE) distributions from each variable selection method conducted on design matrices filtered to enforce a maximum correlation coefficient magnitude between covariate pairs of 0.4 or 0.95 (|r| ≤ 0.4 or |r| ≤ 0.95).
posted on 2016-09-07, 17:40authored byBenjamin R. Fitzpatrick, David W. Lamb, Kerrie Mengersen
The final column contains the coefficient of determination (R2) values for the model-averaged predictions (MAP) from the models resulting from the combinations of variable selection technique and design matrix filtering austerity specified by that row. LAR = Least Angle Regression Variable Selection, Exh = Exhaustive Search Variable Selection, Seq = Sequential Replacement Variable Selection, Fwd = Forward Stepwise Variable Selection, Bwd = Backward Stepwise Variable Selection, Min. = Minimum, 1st Qu. = First Quartile, 3rd Qu. = Third Quartile and Max. = Maximum.