<p>This study investigated the reliability-based stability of a cantilever retaining wall using probabilistic analysis and data-driven modeling. Uncertainty in soil and material properties was incorporated by varying input parameters within realistic ranges, and external failure modes including bearing capacity, overturning, and sliding were evaluated using established geotechnical formulations. Reliability index and probability of failure were estimated using classical reliability methods, and the resulting dataset was used to develop hybrid extreme gradient boosting models optimized with metaheuristic algorithms. Model performance was evaluated using multiple statistical indices on training and testing datasets. Results showed strong agreement between predicted and reference values, which indicated that the models captured nonlinear system behavior with high accuracy. Rank-based evaluation provided balanced model comparison, and distributional consistency was verified using the Anderson–Darling test. Explainable analysis using Shapley additive interpretation quantified the relative importance of input variables and confirmed that friction angle, unit weight, and cohesion governed system response. The proposed framework achieved high predictive accuracy with lower computational demand compared with repeated simulation-based reliability analysis and demonstrated effectiveness for reliable stability assessment of retaining wall systems under uncertainty.</p>