This work proposes the BEmXRD-Nets framework, a novel machine learning framework that integrates fundamental atomic properties with learned embeddings from experimental X-ray diffraction (XRD) patterns to accurately predict the crystal energy in diverse and complex material structures. This framework is implemented to predict both formation and total energy for structures ranging from binary ( \(\text {A}_{k}\text {B}_{l}\) ) to complex quinary ( \(\text {A}_{k}\text {B}_{l}\text {C}_{m}\text {D}_{n}\text {E}_{p}\) ) compositions. The elemental and structural features are extracted to create tailored representations for each crystal type, which are then utilized to train several machine learning models and combined using an optimized-weight stacking ensemble (STE) technique. The numerical results indicate that integrating elemental properties with XRD embeddings significantly enhances prediction accuracy and stability across all configurations. Our framework provides a robust and generalizable method for estimating crystal energies in complex structures, thereby advancing the application of machine learning in the discovery and design of novel materials.
