[17a-S2_204-1~13] 23.1 Joint Session N "Informatics"

11:30 AM - 11:45 AM JST(2:30 AM - 2:45 AM UTC)

[17a-S2_204-10] Screening Superconducting Hydrides by Ranking Estimation Accuracy and Ensemble Predictions

〇Yukito Kimura¹, Souta Miyamoto¹, Katsuaki Tanabe¹ (1.Kyoto Univ.)

Keywords:

superconductor,hydrides,materials informatics

We developed machine-learning models for discovering previously unknown superconducting hydrides using a dataset of binary and ternary hydrides. To reflect composition-level generalization, we used a dataset split that prevents data with the same composition but different pressures from appearing in both the training and validation sets, and we evaluated the models using both prediction-error metrics and ranking-performance metrics. Based on these evaluations, we screened superconducting hydride candidates from the ensemble predictions of multiple models. We are currently validating the selected candidates and further improving model and feature design informed by physicochemical insights.