Toward Practical Rock Mass Classification and Conversion: An Interpretable Data-Driven Framework for Mapping Q-System Parameters to RMR

Abstract Rock mass classification is fundamental to underground engineering design, and accurate conversion between the Q-system and rock mass rating (RMR) is important where the two systems are used together. This study assembles a database of 448 paired Q-system–RMR records from field measurements in Zimbabwe and published tunnel case histories. Statistical reference models are first used to reassess classical conversion formulas. The recalibrated linear relation RMR = 6.70ln( Q ) + 46.31 outperforms the two most widely used traditional formulas, while the spline model reveals marked nonlinearity at low and high Q levels, indicating that a global linear relation is insufficient for reliable conversion. To improve predictive performance, multiple regression models are systematically compared, and the extreme gradient boosting model optimized by the opposition-based gravitational-search algorithm (OGSA–XGBoost) is selected, which achieves coefficients of determination ( R 2 ) of 0.9506 and 0.9382 on the validation and test sets, together with leading hit rates under several tolerance thresholds and residuals dominated by random scatter rather than systematic bias. Interpretation analysis indicates that rock quality designation (RQD) and joint roughness ( Jr ) contribute positively to predicted RMR, whereas joint alteration ( Ja ) and the stress reduction factor (SRF) exert negative effects. Under high stress or weak structure conditions, the adverse effects of Ja and SRF can partially offset the benefits of high RQD and favorable joint roughness. Overall, the results clarify the applicability of classical formulas under the present database and provide an interpretable framework for Q-system to RMR conversion under engineering conditions represented by this database.