Feature-Assisted Machine Learning for Predicting Band Gaps of Binary Semiconductors.

The band gap is a key parameter in semiconductor materials that is essential for advancing optoelectronic device development. Accurately predicting band gaps of materials at low cost is a significant challenge in materials science. Although many machine learning (ML) models for band gap prediction already exist, they often suffer from low interpretability and lack theoretical support from a physical perspective. In this study, we address these challenges by using a combination of traditional ML algorithms and the 'white-box' sure independence screening and sparsifying operator (SISSO) approach. Specifically, we enhance the interpretability and accuracy of band gap predictions for binary semiconductors by integrating the importance rankings of support vector regression (SVR), random forests (RF), and gradient boosting decision trees (GBDT) with SISSO models. Our model uses only the intrinsic features of the constituent elements and their band gaps calculated using the Perdew-Burke-Ernzerhof method, significantly reducing computational demands. We have applied our model to predict the band gaps of 1208 theoretically stable binary compounds. Importantly, the model highlights the critical role of electronegativity in determining material band gaps. This insight not only enriches our understanding of the physical principles underlying band gap prediction but also underscores the potential of our approach in guiding the synthesis of new and valuable semiconductor materials.