System-level reliability evaluations of CRWs under static conditions through SCM and hybrid ensemble learning

This study presents a reliability and system reliability assessment framework for cantilever retaining wall under limit equilibrium conditions using hybrid ensemble-learning techniques integrated with sequential compounding method. Variability in soil and material properties was incorporated using a Latin Hypercube Sampling-Monte Carlo Simulation (LHS-MCS) method, while deterministic formulations are used to generate datasets for training hybrid ensemble-learning model. External failure modes such as sliding, overturning, and bearing capacity are considered across six coefficients of variation (COV) sets. Results indicate that bearing capacity governs dominant failure mode in the system reliability analysis, exhibiting high probabilities of failure (POF = 0.40–0.49) and low reliability indices (β = 0.16–0.33), whereas sliding and overturning show comparatively higher reliability ( \:β\:\:>\:1.5 ). Despite safe component-level responses, system reliability remains low (β = -0.09 to 0.24), demonstrating the importance of system-based evaluation. Among the computational methods, extreme gradient boosting optimized with grey wolf optimizer demonstrates superior performance with coefficient of determinations ( \:R²) values of 0.9741 (training) and 0.9592 (testing) for sliding, 0.9494 and 0.9358 for overturning, and 0.9598 and 0.9465 for bearing capacity, along with the lowest RMSE and highest rank score (15). Furthermore, Anderson darling test, Taylor plot, close agreement between predicted and actual β values confirms model robustness. The findings establish the proposed structure as an efficient and reliable method for probabilistic CRWs stability assessment.