A logistic regression classifier for long-term probabilistic prediction of rock burst hazard

Rock burst is a complex dynamic process can lead to casualties, to failure and deformation of the supporting structures, and to damage of the equipment on site; hence, its prediction is of great importance in underground construction. We present a novel empirical method to predict rock burst based on the theory of logistic regression classifiers. An extensive database collected from the literature, which includes observations about rock burst occurrence (or not) in underground excavations in projects from all over the world, is used to train and validate the model. The proposed approach allows us to compute new class separation lines (or planes) to estimate the probability of rock burst, using different combinations of five possible input parameters—tunnel depth, H; maximum tangential stress, MTS; elastic energy index, Wet; uniaxial compressive strength of rock, UCS; uniaxial tensile strength of rock, UTS—among which it was found that the preferable model could be developed in H–Wet–UCS space. The proposed model is validated with goodness-of-fit tests and nine-fold cross-validation; results show that its predictive capability compares well with previously proposed empirical methods and confirm that, as expected, the probability of rock burst increases with excavation depth, and that both Wet and UCS have a similarly significant influence on rock burst occurrence. Finally, expressions are proposed for identification of conditions associated with several reference values of rock burst probability, which can be employed in preliminary risk analyses.