Exploring modeling techniques for predicting band gaps of Doped-ZnO: A Machine learning approach

Abstract

Machine learning, as one of the promising alternatives for solving complex challenges, has recently received considerable attention. In this study, we apply several well-established machine-learning models for predicting the energy band gap of doped-ZnO as well as novel doping concentrations. This approach significantly expands the possibilities for designing functional materials, offering innovative solutions to meet current energy needs. The results show that the Gaussian Process Regression (GPR) model achieved outstanding performance, with a correlation coefficient (CC) of 98.97%, a root mean square error (RMSE) of 0.0022, and a mean absolute error (MAE) of 0.0020. Comparatively, the Support Vector Machine (SVM) model recorded a CC of 83.70%, an RMSE of 0.0052, and an MAE of 0.0048, while the Random Forest model exhibited a CC of 76.40%, an RMSE of 0.0086, and an MAE of 0.0083. These results underscore the exceptional effectiveness of the GPR model in predicting material properties, while also highlighting the significant contributions of the SVM and Random Forest (RF) methods. This study opens up new research avenues in the fields of materials science and catalysis by exploring the predictive capabilities of different machine learning models for designing functional materials. We emphasize that the selection of the appropriate modeling method is critical for accurately predicting material properties. These results pave the way for future investigations aimed at refining and further comparing the performances of different modeling methods to optimize photocatalytic materials and address the challenges of clean energy.