Abstract:
How to minimize misclassification errors has been the main focus of
Inductive learning techniques, such as CART and C4.5. However, misclassification
error is not the only error in classification problem. Recently, researchers
have begun to consider both test and misclassification costs. Previous works assume
the test cost and the misclassification cost must be defined on the same
cost scale. However, sometimes we may meet difficulty to define the multiple
costs on the same cost scale. In this paper, we address the problem by building
a cost-sensitive decision tree by involving two kinds of cost scales, that minimizes
the one kind of cost and control the other in a given specific budget. Our
work will be useful for many diagnostic tasks involving target cost minimization
and resource consumption for obtaining missing information.
