Machine-Learning-Guided Prediction and Statistical Analysis of Electronic Bandgaps in Double-Perovskite Oxides for Photovoltaic and Optoelectronic Design
Year: 2026
Type: preprint
Abstract: Double-perovskite oxides (A2BB′O6) are attractive for photovoltaic and optoelectronic applications, but their bandgap depends on a compositional space too large for routine DFT screening. In this study, we build composition-only machine-learning models on 1,306 DFTcomputed (GLLB-SC) double perovskites, using 72 descriptors derived from elemental electronegativity, ionic/atomic radius, group and period position, Mendeleev number, and first ionization energy. Six regression algorithms were compared under 10-fold cross-validation where a tuned XGBoost model gave the best results (test R2 = 0.975, MAE = 0.188 eV, RMSE = 0.245 eV), outperforming all others with statistical confidence (Friedman χ2 = 45.14, p =1.4×10-7; Wilcoxon p < 0.005 throughout). SHAP analysis pointed to A-site group number and B-site electronegativity and radius as the main drivers of bandgap, matching the mechanisms reported in the original DFT study. These results demonstrate that interpretable, composition-only machinelearning models can reproduce DFT-level bandgap trends at negligible computational cost, providing a practical high-throughput screening tool for lead-free double-perovskite discovery.
Source: ChemRxiv
Authors: Md Ragib Abid, Md Shafayatul Islam
Language: English
Subfield: Materials Chemistry
Field: Materials Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7223402345