Accelerating Solar Cell Material Discovery Using Machine Learning Models Trained on DFT Databases for Bandgap Prediction and Crystal Structure Stability

The global transition toward renewable energy has placed unprecedented pressure on the photovoltaic (PV) community to discover new solar absorber materials that are efficient, stable, non-toxic and scalable. Traditional Edisonian screening and even Density Functional Theory (DFT) high-throughput calculations remain computationally expensive, often requiring 10⁴–10⁶ CPU-hours to survey modest chemical spaces. This paper presents a Master's-level review and technical analysis of how machine learning (ML) models trained on established DFT databases—including the Materials Project, OQMD, AFLOW and JARVIS-DFT—are transforming the discovery pipeline for next-generation solar cell materials. We focus on two properties central to PV performance: the electronic bandgap (E_g) and thermodynamic/structural stability expressed through the formation energy (E_form) and energy above the convex hull (E_hull). Random Forest, Gradient Boosted Regression, Kernel Ridge Regression, and Graph Neural Networks such as CGCNN and MEGNet are compared with respect to mean absolute error, training cost and transferability. On benchmark datasets, modern GNNs achieve bandgap MAEs of 0.28–0.32 eV and formation energy MAEs below 30 meV/atom, enabling the screening of more than 10⁵ candidate structures in minutes. Applications to halide perovskites, chalcogenides and vehicle-integrated PV (VIPV) are highlighted, along with current limitations involving dataset bias, interpretability and experimental validation loops. Keywords: Bandgap prediction, Crystal structure stability, Density Functional Theory, Machine learning, Materials discovery, Photovoltaics, Solar cells. Title: Accelerating Solar Cell Material Discovery Using Machine Learning Models Trained on DFT Databases for Bandgap Prediction and Crystal Structure Stability Author: Eng. Nawaf F DH Almutairi, Eng. Ali Mejbel Aljadei International Journal of Mathematics and Physical Sciences Research ISSN 2348-5736 (Online) Vol. 14, Issue 1, April 2026 - September 2026 Page No: 11-18 Research Publish Journals Website: www.researchpublish.com Published Date: 21-April-2026 DOI: https://doi.org/10.5281/zenodo.19679862 Paper Download Link (Source) https://www.researchpublish.com/papers/accelerating-solar-cell-material-discovery-using-machine-learning-models-trained-on-dft-databases-for-bandgap-prediction-and-crystal-structure-stability