Back to Search
Start Over
Machine learning regression model for predicting the band gap of multi-elements nonlinear optical crystals.
- Source :
-
Computational Materials Science . Jun2024, Vol. 242, pN.PAG-N.PAG. 1p. - Publication Year :
- 2024
-
Abstract
- [Display omitted] • Combination of classification model and regression model in machine learning. • First principles calculations. • Band gap of nonlinear optical crystals. Nonlinear optical crystals (NLO) are regarded as catalytic materials for photon coupling due to their characteristics of high conversion efficiency, tunability, and ease of manipulation. We aim to find superior performing NLO crystals by precisely estimating their bandgaps. Our approach involves utilizing compositional, structural, and orbital data as input features, leveraging models based on the Random Forest regression (RFR), Extreme Gradient Boosting (XGB), and Gradient boosting regression (GBR) regression algorithms. Notably, the RFR regression model in our research exhibited exceptional predictive performance, attaining an R2 value of 0.823, with an RMSE of 0.639 eV and an MAE of 0.470 eV per atom. Our model works well on a small dataset. Moreover, we incorporated Shapley Additive exPlanations (SHAP) analysis to elucidate the rationale behind predictions by assessing the contribution of each feature to the bandgap. To verify the reliability of our model, we selected three traditional crystals and conducted first-principles electronic structure calculations. We compared the predicted values from our model with those computed using the Heyd-Scuseria-Ernzerhof (HSE) hybrid functional, ensuring an error margin of approximately 0.5 eV, thus confirming the precision of our model. These models facilitate swift and cost-effective predictions of bandgaps in NLO crystals. [ABSTRACT FROM AUTHOR]
Details
- Language :
- English
- ISSN :
- 09270256
- Volume :
- 242
- Database :
- Academic Search Index
- Journal :
- Computational Materials Science
- Publication Type :
- Academic Journal
- Accession number :
- 177483278
- Full Text :
- https://doi.org/10.1016/j.commatsci.2024.113109