Cost-Sensitive Classification with Unconstrained Influence Diagrams

Abstrakt

In this paper, we deal with an enhanced problem of cost-sensitive classification, where not only the cost of misclassification needs to be minimized, but also the total cost of tests and their requirements. To solve this problem, we propose a novel method CS-UID based on the theory of Unconstrained Influence Diagrams (UIDs).

We empirically evaluate and compare CS-UID with an existing algorithm for test-cost sensitive classification (TCSNB) on multiple real-world public referential datasets. We show that CS-UID outperforms TCSNB.

Klíčová slova

• trees
• knowledge
• cost-sensitive classification