An efficient and interpretable method for slope stability assessment and optimisation

Harnessing the power of data-driven techniques for highly nonlinear prediction and multi-objective optimization, we first assembled a 1,080-sample dataset that captures the critical geometric and mechanical drivers of mine-slope stability. An extreme-gradient-boosting regressor (XGBR) was then trained to forecast stability. After hyper-parameter tuning via the cultural algorithm (CA), the CA-XGBR model consistently ranked top on every performance metric. Global interpretability was supplied with SHAP, quantifying each feature’s marginal contribution to the predicted safety factor. Finally, the CA-XGBR—augmented with closed-form stability constraints—was embedded as one objective within a multi-objective framework and solved by NSGA-II. The resulting prediction–optimization platform outperforms conventional limit-equilibrium designs in a benchmark open-pit case, offering a new, fully data-driven paradigm for geotechnical slope assessment and geometry tuning.