Bayesian sparse modeling for interpretable prediction of hydroxide ion conductivity in anion-conductive polymer membranes

Anion-conductive polymer membranes have attracted significant attention as solid electrolytes for alkaline fuel cells and electrolyzers. Their ionic conductivity varies depending on the type and arrangement of quaternary ammonium groups, as well as the structure and connectivity of hydrophilic and hydrophobic domains. In particular, the size and connectivity of hydrophilic domains strongly influence hydroxide ion transport. However, this relationship is only understood at a qualitative level, and it has not been clear which specific features govern conductivity in a quantifiable manner. To address this issue, we employed Bayesian sparse modeling to simultaneously predict conductivity and perform feature selection, thereby constructing an interpretable model. This approach enables robust feature selection even when applied to small datasets, which are common in advanced materials research. As a result, we identified that the ratio of the number of carbon atoms in the hydrophobic to hydrophilic segments is a key feature for predicting the conductivity of anion-conductive polymer membranes in the present dataset.