Rockburst and Coal-Mine Seismic Hazard Prediction Using Empirical Criteria and Machine Learning

Rockbursts and mining-induced seismic bumps injure miners and destroy openings, and mines still rate the hazard with single-indicator empirical criteria. This project rebuilds the classical rockburst criteria (Russenes, Hoek, Turchaninov, rock brittleness, the strain energy index Wet) and compares them with machine learning (Random Forest, XGBoost, LightGBM, SVM, logistic regression, with and without SMOTE) on a real 239-case rockburst database under an honest duplicate-grouped, repeated cross-validation. The comparison is repeated on real shift-by-shift monitoring data of a Polish coal mine (UCI seismic-bumps) with a time-ordered split, against the mine's own seismic and seismoacoustic hazard ratings. SHAP and partial dependence explain the models, leakage traps are demonstrated, and a web application offers a rockburst risk calculator and a shift hazard monitor with alarms on five databases. Weka and GNU Octave re-check the results; a complete MATLAB version is included.

Category: Mining Engineering

Mining Engineering Project MN03 from ProTechSC Academy, with graphical abstract, video demonstration and project guidance.