Slope Stability Classification with XGBoost Optimized by an Improved Remora Optimization Algorithm

Rapid and accurate slope-stability prediction is important for engineering safety. However, non-linear interactions among multiple influencing factors complicate slope-stability prediction, limiting the effectiveness of conventional methods under complex slope conditions. To address these issues, an Improved Remora Optimization Algorithm (IROA) is developed and used to optimize XGBoost hyperparameters for slope-stability classification. IROA integrates hybrid Latin hypercube–random initialization, probabilistic multistrategy search, and stagnation-triggered recovery. A 23 full-factorial ablation study on CEC2017 F16 showed that complete IROA and the recovery-only variant reduced the mean objective value by 24.03% and 22.91%, respectively, relative to standard ROA. At D = 50, IROA achieved the lowest mean objective value on all 12 selected CEC2017 functions, with an average rank of 1.000 compared with 2.500 for the next-best algorithm, BWO. Using 256 slope cases and six input variables—unit weight, cohesion, internal friction angle, slope angle, slope height, and pore-pressure ratio—XGBoost hyperparameters were selected through five-fold cross-validation within the training set. On the held-out test set, IROA–XGBoost achieved an accuracy of 0.9608, an F1 score of 0.9697, and an area under the receiver operating characteristic curve (AUC) of 0.9411. Based on the reported accuracy values, observed test accuracy was 11.77 and 3.92 percentage points higher than that of baseline XGBoost and ROA–XGBoost, respectively. Gain-based feature-importance analysis was used to interpret the relative contributions of the inputs within the fitted model. Predictions for all ten supplementary engineering cases agreed with their source-reported stability states. These results suggest that IROA–XGBoost is a promising approach to slope-stability classification and may support engineering assessment under the evaluated conditions.