Hybrid Feature Learning for Crystal System and Space Group Classification from Standard XRD Patterns

The ability to identify crystal structures correctly is essential for predicting material performance and the creation of novel materials. As the internal order of the materials, XRD (x-ray diffraction) patterns serve as a crucial tool for structure determination. On the other hand, conventional XRD analysis approaches largely depend on professional experience and manual peak matching, which leads to low efficiency and susceptibility to subjective errors. Machine learning approaches have recently shown enormous promise for structural analysis, although studies that currently exist are frequently restricted to simple classification tasks, which still have difficulties with fine-grained tasks like space group recognition and strong performance when there is noise. In this paper, we present a hybrid feature learning method that uses a LightGBM classifier to identify crystal systems and space groups based on large-scale simulated XRD patterns created from the International Crystal Structure Database (ICSD) to combine handcrafted physical features with deep convolutional features in order to identify crystal systems. Extensive research shows that the suggested approach outperforms conventional machine learning models by achieving 97.8 % accuracy in the classification of the crystal system and 87.2 % accuracy in space group categorization. Robustness tests under noise and peak shifts validate its stability. These results suggest that integrating domain knowledge and data-driven features enhances intelligent XRD analysis, offering a promising strategy for high-throughput material discovery.