Fig. 3
Decision tree constructed from the trained model with the highest accuracy of the XGBoost. In OSDA-free synthesis of zeolites, the most significant synthesis descriptors for zeolite phase selection are the amounts of SiO2, Al2O3, MOH (M = Li, Na, K, etc.), and H2O present in the synthesis mixture. Machine learning models including XGBoost, support vector machine, decision tree, and random forest were trained to predict synthesis results from synthesis descriptors including temperature, heating time, and chemical compositions with different standard denominators. The trained model with the highest accuracy was the XGBoost model using (Si + Al) as the denominator, and this model was interpreted as a decision tree shown here with a depth of 4. The complete tree (depth = 12) can be found in Supplementary Figs. 5–11. The most dominant crystalline phases in the predictions are presented. The percentages represent the fractions that the dominant phases appear in the deeper branches in the complete tree
