Highly correlated variables; maximum elevation with elevation range (r = 0.99), and mean annual temperature with latitude (r = −0.92), substituted for each other in the top models, although latitude (0.52) was considerably higher in relative variable importance than mean annual temperature (0.20).
Initially an optimal set of 20 genes was selected by removing redundant probe sets and extracting the top 100 genes (by reported Gini variable importance), k-means clustering (k = 20) these genes and selecting the best gene from each cluster (again by variable importance).
A working value of T# = 300 was chosen for the RFtb model structure used in the tests and experiments Fig. 9 variable importance calculated by mean square error (MSE) and purity or entropy degree.
For RF, we considered the mean decrease in accuracy to assess variable importance.
One of the measures of variable importance is the mean decrease in accuracy, calculated using the out-of-bag sample.
Mean Decrease Gini (MDG) served as a variable importance measure (VIM) for our study as it was shown to be more robust in previous research [ 39].
Practically, this approach counts variable importance by calculating the average mean squared error (MSE) provided by RF from a series of runs as a tool to rank the predictors.
Finally, we used the variable importance scores, percent change in mean squared error (MSE), to identify loci that had the greatest influence on classification, by choosing those loci whose percent change in MSE was greater than 5%.
Variable importance measures were implemented through mean decrease in accuracy and the Gini Index (GI) [ 24], to find the genes that best discriminate between the different disease phenotypes.
To determine which variables (biochemicals) made the largest contribution to the classification, a variable importance measure was computed, termed the mean decrease accuracy (MDA).
