测试时间代价敏感决策树

Testing time cost-sensitive decision trees

决策树作为一种重要的分类算法已在许多领域得到了广泛应用。针对传统决策树算法未考虑实际应用中代价约束的问题,已有研究提出在限制代价的情况下构建决策树的方法。这些方法中代价的定义并没有考虑测试样本通过决策树进行分类的时间代价。为了最小化样本通过决策树进行分类的测试时间,提出了一种测试时间代价敏感决策树算法。定义了样本的测试时间代价,定义了衡量属性重要度的决策指数,给出了构造代价敏感决策树的算法。实验结果表明,算法的测试时间代价相较于C4.5、RSDT和CSGR等主要算法平均提升了11.7%,且在不同数据集下分类准确度平均提升了5.3%。

As an important classification algorithm,decision trees have been widely used in many fields.Traditional decision tree algorithms,however,do not account for cost constraints in practical applications.Previous research has introduced methods for constructing decision trees under limited cost conditions,but these methods do not consider the time cost associated with testing samples during the classification process.To minimize the testing time for classifying samples using decision trees,this paper proposes a testing time cost-sensitive decision tree algorithm.It defines the test time cost for samples and introduces a decision index to measure attribute importance.The algorithm for constructing a cost-sensitive decision tree is presented.Experimental results show that the proposed algorithm reduces test-time cost by an average of 11.7%compared to major algorithms such as C4.5,RSDT,and CSGR,while also improving classification accuracy by an average of 5.3%across different datasets.