Subscribe Bookmark RSS Feed

Bootstrap forest: measures of variable importance?


Community Member


Oct 7, 2013

Hi All!

I am running my first bootstrap forest analysis in JMP and would really appreciate some suggestions on defining the importance of individual variables.  In the Column Contributions section of the results report, there are two metrics for each predictor variable: "number of splits" and "SS", but I am unsure of their use.  I know that larger bars are associated with more important variables, but is it possible to determine the increase in mean square error that would occur with the removal of any given variable?  I was not able to find detailed information about numeric measures of variable importance in the JMP Help files.  Here's a quick description of my data set...

--270 observations, 9 predictor variables

--one categorical predictor variable, eight continuous predictor variables

--response variable (i.e. the one to be predicted) is continuous

Does anyone have advice, articles, or websites that would help make this clearer?  Thanks!