Multi-strategy quantum-inspired dung beetle optimizer for kernel extreme learning machine in slope safety factor prediction

Accurate prediction of the slope safety factor (FS) is crucial for stability assessment and risk-informed decision making in mining and geotechnical engineering, yet practical applications are often constrained by small, heterogeneous datasets and the sensitivity of learning models to hyperparameter settings. This study presents a quantitative FS prediction framework that couples a Kernel Extreme Learning Machine (KELM) with a multi-strategy quantum-inspired dung beetle optimizer (QHDBO) to tune the key hyperparameters (regularization coefficient and kernel parameter) in logarithmic space. Five widely available geotechnical and geometric variables—unit weight (γ), cohesion (c), internal friction angle (φ), slope angle (α), and slope height (h)—are used as inputs, with FS as the target output. To ensure reliable generalization estimates under limited samples, model performance is assessed using strict no-leakage repeated nested cross-validation (5-fold outer CV repeated 30 times) and evaluated by R², RMSE, MAE, and MAPE, complemented by engineering-critical error analysis near FS ≈ 1 and bootstrap confidence intervals. Results show that QHDBO-KELM achieves superior accuracy and robustness compared with baseline KELM and several benchmark models (RF, GBDT, MLP, SVR), yielding lower prediction errors and higher goodness-of-fit across diverse data partitions. Paired significance tests over 150 outer folds further confirm that the improvements are statistically significant, indicating that the proposed framework provides an efficient and dependable surrogate tool for slope FS prediction and can be extended to other geotechnical stability-related regression tasks.