Identifying domains of applicability of machine learning models for materials science

Published in Nature Communications, 2020

Assessing the performance of machine learning models via their average performance across some unbounded domain can both over- and undersell their merits. Using rule learning (or subgroup discovery) techniques allows to identify an intepretable bounded region where a model behaves execptionally well.

Recommended citation: C Sutton, M Boley, LM Ghringhelli, M Rupp, J Vreeken, M Scheffler. (2020). "Identifying domains of applicability of machine learning models for materials science." Nature Communications. 11(1),4428.
Download Paper