• Uniaxial loading-unloading tests on sandstone with varying grain sizes reveal that existing rockburst tendency criteria may not be universally applicable.
• Integration of ejection kinetic energy with unloading ratio, failure load, WET, and PES significantly improves rockburst tendency classification accuracy.
• Machine learning models, particularly Random Forest for classification and AdaBoost Regressor for regression, provide robust predictions of rockburst tendency.
• The study proposes a novel laboratory-scale approach combining experimental parameters and machine learning for reliable rockburst tendency assessment.