Geek Answers Handbook

Multi-level apprentice solutions

There are many one-dimensional implementations of a student decision tree (C4.5, etc.), but does anyone really know multidimensional decision tree recognition algorithms?

+5
source share
3 answers

Bennett and Blue Support for the vector machine approach to decision trees makes multi-dimensional splits using the built-in SVM for each decision in the tree.

Similarly, in the Multicategory Classification using Discrete Vector Support Machines (2009) , Orsenigo and Vercellis implement a multi-category version of Discrete Vector Support Machines (DSVM) in the nodes of the decision tree.

+6
source

The CART algorithm for the decision tree can be turned into multidimensional. CART is a binary splitting algorithm, not C4.5, that creates a node for a unique value for discrete values. They use the same algorithm for MARS as for missing values.

, node, , , , (, ), node, . , ( ) 1. < 1.0 .. , . node, . . , .

+1

Yes, there are some, such as OC1, but they are less common than those that do unidirectional splits. The addition of multidimensional splits greatly expands the search space. As a kind of compromise, I saw some logical students who simply calculate linear discriminant functions and add them to the list of candidate variables.

0
source

More articles:


All Articles