Feature Selection with Time Cost Constraint

Abstract

Feature selection is an important task in machine learning and data mining. In real world applications, time and money are required to obtain features of objects. Many existing works have been developed to preserve enough information of a decision system, and at the same time, minimize time cost or money cost. In this paper, we study feature selection with time cost constraint. Here time cost consists of testing time cost and waiting cost. The optimization objective is to obtain a feature subset with the lowest conditional informational entropy. Two algorithms are designed to obtain optimal solutions for the problem. We revise the first algorithm by utilizing the relationship of the conditional entropy between a set and its subsets, then obtain the second algorithm. Experimental results on four UCI datasets indicate that the efficiency improvement is significant.