Cost-Time Sensitive Decision Tree with Missing Values

Abstract

Cost-sensitive decision tree learning is very important and popular in machine learning and data mining community. There are many literatures focusing on misclassification cost and test cost at present. In real world application, however, the issue of time-sensitive should be considered in costsensitive learning. In this paper, we regard the cost of time-sensitive in costsensitive learning as waiting cost (referred to WC), a novelty splitting criterion is proposed for constructing cost-time sensitive (denoted as CTS) decision tree for maximal decrease the intangible cost. And then, a hybrid test strategy that combines the sequential test with the batch test strategies is adopted in CTS learning. Finally, extensive experiments show that our algorithm outperforms the other ones with respect to decrease in misclassification cost.