A high-throughput experiment and machine learning-integrated approach for rapid delineation of the glass-forming range: a case study of the B2O3–La2O3–BaO system

• High-throughput experiments combined with machine learning for glass-forming region mapping. • RF model accurately predicts glass-forming region with small datasets. • Experimental validation confirms ±2 mol% boundary prediction reliability. • SHAP analysis reveals opposite roles of B₂O₃ and La₂O₃ in glass formation. • A data-driven strategy replaces trial-and-error experiments to support glass composition design. Accurate and efficient determination of the glass-forming region is a fundamental prerequisite for the design and performance optimization of novel glass systems. However, conventional trial-and-error approaches remain time-consuming and costly. In this work, a rapid method for identifying glass-forming regions is developed by integrating high-throughput experiments with machine learning, and the approach is validated using the well-studied B₂O₃–La₂O₃–BaO ternary system. High-throughput melt-quenching experiments were first employed to identify the approximate glass-forming region, and the resulting experimental data were used as a training dataset to develop six machine learning classification models, including Support Vector Machine, Random Forest, Extreme Gradient Boosting, Gaussian Process Classification, Artificial Neural Network, and K-nearest Neighbors. A systematic comparison of model performance indicates that the random forest model exhibits the best overall predictive capability. To assess the reliability of the predicted glass-forming boundary, twenty compositions located within ±2 mol% of the predicted boundary were selected for experimental validation, showing good agreement between model predictions and experimental results and confirming the effectiveness of the proposed method for rapid delineation of glass-forming regions. SHapley Additive exPlanations (SHAP) interpretability analysis was further employed to interpret the prediction results of the random forest model. The results indicate that La₂O₃ exerts a strong inhibitory effect on glass formation, whereas B₂O₃ shows a pronounced promoting effect. This study not only provides a reliable reference for compositional design in the B₂O₃–La₂O₃–BaO system, but also demonstrates the potential of data-driven approaches to accelerate the exploration and development of complex glass materials.