[16a-K505-1~9] 23.1 Joint Session N "Informatics"

10:30 AM - 10:45 AM JST(1:30 AM - 1:45 AM UTC)Presentation by Applicant for JSAP Young Scientists Presentation Award

[16a-K505-6] Ensemble learning for experimental band gap prediction from chemical compositions using neural networks

〇Taichi Masuda¹, Katsuaki Tanabe¹ (1.Kyoto Univ.)

Keywords:

semiconductor,band gap,machine learning

The band gap is a critical physical property that determines the characteristics of electronic and optical devices in semiconductor materials. However, conventional computational methods have faced challenges in terms of accuracy and computational cost. In this study, we propose an ensemble learning model that integrates neural networks to predict semiconductor band gaps with high accuracy and efficiency based solely on elemental compositions. Our developed model, incorporating MPNN and CGAN, demonstrated a mean absolute error of 0.348 eV against experimental values, achieving the highest prediction accuracy among existing machine learning models for experimental band gap prediction. Analysis from the perspectives of bias, variance, and Shapley values demonstrated that MPNN played a crucial role in the ensemble learning predictions.