Quantifying 3D and suction-induced effects on soil slope stability during rapid drawdown: a sensitivity study using the MARS-WOA approach

The study presents a new hybrid model, called MARS-WOA, which predicts the impact of three-dimensional (3D) and suction-induced effects on soil slope stability. The MARS-WOA model combines the Multivariate Adaptive Regression Spline (MARS) with the Whale Optimization Algorithm (WOA) and applies it to four slope stability datasets. These include 2D and 3D datasets to evaluate the 3D effect in saturated soil slopes and NS (no suction) and WS (with suction) datasets to assess the suction-induced effect in unsaturated soil slopes. The MARS-WOA model demonstrated superior predictive modeling capability and performance compared to two other machine learning models, Support Vector Regression (SVR) and Ensemble Boosting Trees (EBT). This was evidenced by the impressively low Root Mean Squared Error (RMSE ≤ 0.04472) and high R-squared (R² ≥ 0.93) values achieved by the MARS-WOA model across all scenarios. The relative importance analysis indicates that the ratio B/H, representing the 3D effect, moderately influences slope stability design, with a relative importance (RI) value of 15.41%. Similarly, the ratio δαn, which indicates the suction-induced effect, moderately contributes to the slope stability model, with an RI value of 15.73%. These findings suggest that the MARS-WOA model is valuable for soil slope stability analysis and design researchers. The model provides valuable insights into the critical factors affecting slope stability, enabling the creation of more dependable slope designs.