A measure of variable importance was estimated by taking a scaled composite of the importance scores outputted by random forests.
In particular, to define solution performance scattering, the fuzzy entropy is used as a global measure of variable dispersion.
Second, we identified the most important metrics based on the Model Improvement Ratio (MIR), a standardized measure of variable importance [51, 52].
The ratio of nonsynonymous to synonymous substitutions, dN/dS, which is a measure of variable Darwinian selective pressures, was estimated by maximum likelihood from the PEG11 coding sequence alignment using the codeml program from PAMLv4 [49].
Class-specific measures of variable importance are also provided in Table 2. ⋆denote significant measure of variable importance in the PLS case, and upper script numbers indicate on which PLS component a variable was found significant at the 5% level.
In addition, a built-in permutation-based measure of variable contribution to prediction accuracy allows to rank variables by their importance.
Classifiers were optimized on a training sample, and they were used subsequently to rank the 49 items according to a permuted measure of variable importance.
The second measure of variable importance uses the total decrease in node impurities derived from splitting on the variable, averaged over all trees and measured by residual sum of squares.
The measure of variable importance in boosted CARTs is based on the number of times a variable is involved in a split, weighted by the squared improvement of the model as a result of the split.
The contribution of our work, however, lies in the discovery that the most naive measure of variable importance in RF, the variable selection frequency (RFSF), actually performs much better than the more popular permutation importance (RFPI) in this context.
