A Robust profit measure for binary classification model evaluation

Franco Garrido, Wouter Verbeke, Cristián Bravo

Abstract: Highlights•We present a measure for profit-driven evaluation of models in the presence of strong variability.•This variability may come from fixed effects, fixed distributions, and random shocks.•The measure was tested both in a synthetic case and an empirical case.•The measure outperforms other commonly used ones in highly variable environments.•The measure allows selecting the most profitable model in the long run.