Two researchers at Duke University have recently devised a useful approach to examine how essential certain variables are for increasing the reliability/accuracy of predictive models. Research could ultimately aid the development of more reliable and better performing machine-learning algorithms
