Interpretable Machine Learning for Predicting Ionic Conductivity of Pore-Filling Membranes

Abstract

Ion exchange membranes (IEMs) are essential components in fuel cells and water electrolysis, and they are central to enabling hydrogen-based energy technologies. Despite their importance, broader commercialization has been slowed by the vast structural diversity of ion-conducting polymers and the labor-intensive, time-consuming nature of membrane characterization, which together demand substantial research and development resources. Although machine learning (ML) has transformed R&D workflows in many disciplines, its adoption in IEM research remains comparatively limited, largely because of scarce and heterogeneous datasets, complex structure–property relationships, and challenges in model interpretability. In this work, we present an ML framework for predicting the ionic conductivity of pore-filling anion exchange membranes (PFAEMs). The model achieves strong predictive performance with low error, suggesting a practical route to shortening development cycles and reducing experimental cost. In addition, feature-level analyses identify the variables that most strongly govern conductivity, offering actionable insight for membrane and polymer design. Overall, this work demonstrates how ML-enabled modeling can improve the efficiency and reliability of anion exchange membrane research and support the development of next-generation ion-conducting materials. Acknowledgments This work was supported This work was supported in part by H2GATHER Program through the National Research Foundation of Korea (NRF) funded by the Ministry of Science and ICT (RS-2025-02303249) and by the National Research Council of Science and Technology (NST) grant funded by the Ministry of Science and ICT (MSIT) of Korea (No. GTL25081-301).