Cost-sensitive Decision Tree with Missing Values and Multiple Cost Scales

Abstract

Most researches focus on two costs for building cost-sensitive decision trees, such as, misclassification costs, test costs. And the existing literatures always consider the two costs as the same scales, for instance, dollars. However, in real application, it is difficult for us to regard two costs as same scales, for instance, considering misclassification cost as a dollar unit. In this paper, a new splitting attributes criterion which is combined with classification ability, test costs and misclassification costs, is proposed under the assumption of multiple-costs scales and with missing values in the dataset. The experimental results show the proposed method outperforms the existed methods in terms of the decrease of misclassification cost.