Practical Predictive Analytics

Random forests

Note that Random Forests (tm) is a trademark of Leo Breiman and Adele Cutler and is licensed exclusively to Salford Systems for the commercial release of the software.

One problem with decision trees is that they are highly dependent on the specific variables that are initially chosen. One could generate a slightly different set of initial variables and end up with an equally valid solution, although different.

Random forest algorithms (which are known by several different names refer to the note at end of the chapter) are an attempt to improve decision trees in that random forests randomize the selection of variables and subsamples in order to generate multiple (even thousands of) separate decision trees. After all the decision ...