Multiple-scale cost sensitive decision tree learning

The previous cost-sensitive learning methods assume that the (medical) test cost is measured in the same scale with the misclassification cost while minimizing the expected total cost. This paper proposes a general target-resource framework involving multiple kinds of cost scales, which minimize one kind of cost scale (called target cost scale) through controlling the others (called resource cost scales) in given resource budgets. The proposed cost-sensitive learning model also assists in, such as healthcare data classification and bioinformatics analysis, which are practical and desired application for developing a multiple-scale cost-sensitive learning tool. We experimentally evaluated our approach using the biological and medical datasets, and demonstrated that our proposed method worked well on learning decision tree under a given budget.