Fast and Frugal heuristics augmented: When machine learning quantifies Bayesian uncertainty

Abstract

Heuristics aim at providing good and fast approximations to complex optimal solutions. They are conceptually simple, implementing them rarely requires high levels of mathematical sophistication or even programming skills. For instance, Fast and Frugal Trees are very simple decision trees for binary classification problems. They are fast and frugal as they rely on a minimum of time, knowledge and computation to make efficient decisions. These advantages come at a cost as well. Their intrinsic nature prevents them from evaluating the accuracy of their estimation. On the opposite, machine learning methods are now widely used to assess predictive Bayesian uncertainty. This article combines the best of the two worlds by introducing a two-step decision making process that combines the simplicity of an heuristic driven tree with a Bayesian estimation of uncertainty. In short, we argue that one should use intuition to form hypotheses, apply statistics to consolidate them (i.e. the Fast and Frugal Tree) and more complex algorithms to estimate their predictive capacities. We apply our methodology to data on loan approval/denial decisions.