Decision trees with minimal costs

Hybrid Cost-Sensitive Decision Tree

“…As pointed out by [14] it is obviously an oversight. As far as we know, the only work considering both misclassification and test costs includes [13,15,9,2]. We discuss these works in detail below.…”

“…Ling et al [9] propose a cost-sensitive decision tree learning program that minimizes the total cost of tests and misclassifications. They also propose several test strategies, and compare their results to C4.5.…”

“…Ling, et al [9] described a simple strategy called Optimal Sequential Test (or OST in short) that directly utilizes the decision tree built to guide the sequence of tests to be performed in the following way: when the test example is classified by the tree, and is stopped by an attribute whose value is unknown, a test of that attribute is made at a cost. This process continues until the test case reaches a leaf of the tree.…”

“…In [9] a very simple heuristic is described. The basic idea is that when a test example is classified by a minimum-cost tree and is stopped by the first attribute whose value is unknown in the test case, all unknown attributes under and including this first attribute would be tested, as a single batch.…”

Cost-Sensitive Decision Trees with Pre-pruning

“…The first category is concerned with making the attribute splitting criterion sensitive to cost [11,12,13]. The other category develops new or modified pruning algorithms to minimize the expected cost [14,15].…”