Simple Test Strategies for Cost-Sensitive Decision Trees

We study cost-sensitive learning of decision trees that incorporate both test costs and misclassification costs. In particular, we first propose a lazy decision tree learning that minimizes the total cost of tests and misclassifications. Then assuming test examples may contain unknown attributes whose values can be obtained at a cost (the test cost), we design several novel test strategies which attempt to minimize the total cost of tests and misclassifications for each test example. We empirically evaluate our treebuilding and various test strategies, and show that they are very effective. Our results can be readily applied to real-world diagnosis tasks, such as medical diagnosis where doctors must try to determine what tests (e.g., blood tests) should be ordered for a patient to minimize the total cost of tests and misclassifications (misdiagnosis). A case study on heart disease is given throughout the paper.