Machine‐Learning Guided Prediction of Band Edge Energies of Halide Double Perovskites for Design of Efficient Perovskite Solar Cells

The performance of perovskite solar cells critically depends on the accurate determination of band‐edge energies, particularly the valence band maximum (VBM) and conduction band minimum (CBM), which govern charge transport and interfacial energy‐level alignment. However, conventional first‐principles methods such as density functional theory (DFT) are computationally expensive and hinder large‐scale exploration of the vast compositional space of halide double perovskites. In this work, we develop a machine‐learning‐driven high‐throughput framework for predicting the band‐edge energies of A 2 B′B″X 6 halide double perovskites using composition‐based feature descriptors. Multiple supervised learning models were systematically evaluated alongside ensemble approaches across different train–test splits. Among them, the stacking ensemble model with an 85:15 split achieved the best predictive performance, with R 2 values of 0.8540 for VBM and 0.8619 for CBM. The trained model was subsequently applied to a combinatorial library of 5,418 candidate compositions, which was reduced to structurally feasible materials through geometric screening. Further environmental, economic, and band gap filtering yielded 4 promising lead‐free double perovskite candidates suitable for photovoltaic applications. Overall, this work establishes a scalable and physically interpretable machine‐learning framework for rapid band‐edge prediction, significantly reducing reliance on computationally expensive DFT calculations and enabling efficient high‐throughput discovery of sustainable next‐generation photovoltaic materials.