Abstract: Cost-sensitive decision trees usually concern the discussion of the test cost and misclassification cost. During the classification process, splitting attribute selection is the most important. The paper analyzed the disadvantages and the advantages of the existing methods and proposed a novel method that combined the information ratio in information theory with the cost including the test cost and the misclassification cost to select the split attributes. The experimental results show that this method outperforms significantly the existing methods.

Key words: cost sensitive, decision tree, splitting attribute

摘要： 代价敏感决策树通常讨论测试代价和误分类代价，在其分类过程中，最关键的是节点分裂属性的选择。分析了代价敏感决策树分类问题目前常见的选择分裂属性方法的优、缺点，提出了综合信息量和测试代价并且最大程度降低误分类代价的分裂属性选择方法，UCI数据集实验结果显示该方法在各个方面好于已有的方法。

关键词: 代价敏感, 决策树, 分裂属性