Practical Predictive Analytics

Disadvantages of decision trees

Most decision trees utilize what are known as greedy algorithms. There are many definitions of what greedy means, but I like the one that suggests that it always makes the optimal choice based upon whatever information happens to be available at that moment. There is no hindsight that is used when building trees. Slight changes in data can alter results, and if you run the algorithm on a subsequent sample, the results may not replicate in exactly the same way. It is also difficult to force the tree to split the way you want it to when running in an automated fashion. If you want to control splits exactly, you may be better off using interactive decision trees in which you are able to split a node by pointing ...