Database Reference
In-Depth Information
Cost of Misclassification Errors — specifies the cost of assigning a case
to class i , when it actually belongs in class j .
Cost of Tests — specifies the cost of using a certain attribute. For
instance, in medical application, doing an X-Ray has a cost.
Cost of Teacher — the cost associated with providing classified examples
(training set) to the inducer.
Cost of Intervention — This is the cost required to change the value of
an attribute in order to gain more profits.
Cost of Computation — The cost associated with using computers for
training a classifier or testing it.
Human-Computer Interaction Cost — specifies the cost of a user to
employ a learning algorithm.
12.3 Learning with Costs
Individually making each classification learner cost-sensitive is laborious,
and often non-trivial. A framework that can be used to configure new deci-
sion tree models based on a combination of several factors (e.g. accuracy,
measurement cost, misclassification cost, achievement cost) is presented
in [ Someren et al . (1997) ] . On the other hand there are general methods
for making accuracy based classifiers cost-sensitive. [ Domingos (1999);
Zadrozny and Elkan (2001) ] . Nonetheless, these cost sensitive methods
where inseparable and inherent to the concept learning models. Almost all
such (accuracy-based and cost-based) learning models assume the passive
role of predicting class membership by producing accurate classifiers.
Few papers investigated different approaches to learning or revising
classification procedures that attempt to reduce the cost of misclassification.
The cost of misclassifying an example is a function of the predicted class
and actual class represented as the cost matrix C [Pazzani et al . (1994)]:
C ( predictedClass, actualClass )
(12.1)
where C ( P, N ) is the cost of a false positive, and C ( N, P )isthecostof
a false negative misclassification, then the misclassification cost can be
calculated as:
Cost = FP
·
C ( P, N )+ FN
·
C ( N, P )
(12.2)
The cost matrix is an additional input to the learning procedure and
can also be used to evaluate the ability of the learning program to reduce
misclassification costs. Cost may have any units; the cost matrix reflects
the intuition that it is more costly to underestimate how ill someone is
Search WWH ::




Custom Search