Exploring the classification of painted pottery artifacts from different cultural sites in Northwest China by using LIBS technology combined with machine learning

• Combined LIBS technology with machine learning to classify painted pottery artifacts. • Introduced a non-destructive analytical approach to enhance archaeological studies. • Achieved over 95 % accuracy using LightGBM with enhanced feature selection methods. • Provides a scalable, efficient framework for cultural heritage artifact classification. Painted potterys, as a significant aspect of material culture, provides valuable insights into ancient societies and is essential for studying the evolution of civilizations. Traditional identification methods for the age of painted pottery artifacts often rely on excavation sites and decorative features, depending on expert experience, which introduces the risk of subjective misjudgment. To complement these approaches, this study proposes an auxiliary technique using Laser-Induced Breakdown Spectroscopy (LIBS) and machine learning. We analyzed pottery from six Neolithic to Bronze Age sites (c. 3300–600 B.C.). Hierarchical clustering evaluated chemical similarities, while Convolutional Neural Networks (CNN), Deep Neural Networks (DNN), and Light Gradient Boosting Machine (LightGBM) models were employed for classification. The optimal model achieved 98 % accuracy. The LIME interpreter further identified significant elemental features for each culture, enhancing objectivity. This research demonstrates that LIBS with machine learning serves as a valuable complement to traditional expertise, mitigating subjective risks and providing a reliable analytical aid for pottery classifications.