Introduction to Decision Trees - Data Mining with Decision Trees: Theory and Applications - page 11

YRSJOB

<2

2 ≥

LT V

MARST

<75%

75% ≥

YRSADD

DEPEND

M

D

A

>0

<1.5 =0 ≥

1.5

D

A

A M

Fig. 1.4 Underwriting decision tree.

(PAYINC), interest rate (RATE), years at current address (YRSADD),
and years at current job (YRSJOB).
Based on the above information, the underwriter will decide if the
application should be approved for a mortgage. More specifically, this
decision tree classifies mortgage applications into one of the following two
classes:

Approved (denoted as “A”) — The application should be approved.

•

Denied (denoted as “D”) — The application should be denied.

•

Manual underwriting (denoted as “M”) — An underwriter should
manually examine the application and decide if it should be approved (in
some cases after requesting additional information from the applicant).
The decision tree is based on the fields that appear in the mortgage
application forms.

•

The above example illustrates how a decision tree can be used to
represent a classification model. In fact, it can be seen as an expert system
which partially automates the underwriting process. Moreover, the decision
tree can be regarded as an expert system which has been built manually by
a knowledge engineer after interrogating an experienced underwriter at the
company. This sort of expert interrogation is called knowledge elicitation
namely, obtaining knowledge from a human expert (or human experts) to