Measuring real-time predictive models

In this paper we examine the problem of comparing real-time predictive models and propose a number of measures for selecting the best model, based on a combination of accuracy, timeliness, and cost. We apply the measure to the real-time attrition problem.