Databases Reference
In-Depth Information
Buyer #Obs
Y 5
N5
SpeedingCitations
<2
>2
Buyer
# Obs
Buyer
# Obs
Y
1
Y
4
N3
N2
SpeedingCitations
MaritalStatus
S
<0
>
0
M
Buyer
# Obs
Buyer
# Obs
Buyer
# Obs
Buyer
# Obs
Y
0
Y
1
Y
1
Y
3
N2
N1
N0
N2
Figure 4.2
Decision Tree after Splits
Using decision trees
To make a prediction using input data without a corresponding output class, find
the leaf (terminating) node in the tree matching the input values. For example,
suppose that in the example above, we have a single, male showroom visitor
with three speeding citations. The leaf node for this person is the lower right
node of Figure 4.2. The predicted class value is the most frequently occurring
class in the node. The probability of the predicted value is the fraction of that
class within the node. For the example, the predicted Buyer class is Y (3 versus
2) and the probability is 0.6.
Decision tree advantages
The main advantage of decision tree classifiers over other methodologies is that
they are very easy to understand and interpret. The input attribute determining
the first split is, with respect to the full dataset, the most discriminating of the
input attributes. In general, splitting attributes at the top of the tree are more
important than those lower in the tree.
Another advantage of decision trees is that in construction and in application,
they do not require a lot of CPU processing power. Using decision trees to make
classifications is especially fast and simple. If necessary, for a person having
access to the tree structure, the classifications can be made manually in very
little time.
