Advanced Decision Trees - Data Mining with Decision Trees: Theory and Applications - page 171

YRSJOB

<2

≥ 2

LT V

DEPEND

<75% ≥ 75%

>0

=0

A M

A M

YRSJOB

≥ 2

<2

LT V

MARST

<75%

≥ 75%

M

A

D

D

A

Fig. 11.4 Illustration of two regular trees which are equivalent to the option tree
presented in Figure 11.3.

presented in Figure 11.3 can be seen as combining the two regular decision
trees presented in Figure 11.4.
TDDTOP is a Top-Down Decision-Tree (TDDT) inducer which is quite
similar to C4.5 but with the additional ability to create option nodes
[ Kohavi and Kunz (1997) ] . The principal modication to the basic TDDT
algorithm is that instead of always selecting a single test, when several tests
evaluate close to the best test, the TDDTOP algorithm creates an option
node. All the data is sent to each child of an option node, which then splits
the data according to its predetermined test. As for the pruning phase, the
C4.5 pruning algorithm was modified so that the pessimistic error of an
option node was the average of the pessimistic errors of its children.