Answers are grounded in this paper and cite their sources.
The three-dimensional corner effect produces pronounced spatial variation in retaining-wall deformation that cannot be adequately represented by plane-strain analyses or predictions at a single section. This study develops an interpretable machine-learning framework for predicting the normalized maximum wall-deflection envelope of braced excavations in clay. A database of 451 position-wise samples derived from three-dimensional finite-element analyses was characterized using clay type, excavation depth, excavation length-to-width ratio, wall-stiffness coefficient, and normalized wall position. Six regression algorithms were screened, the competitive ensemble models were tuned using the Spider Wasp Optimizer, and the final SWO-XGBoost model was assessed using an inner-outer five-fold out-of-fold procedure. The pooled predictions yielded an R2 of 0.996803, an RMSE of 0.020323, an MAE of 0.012175, a VAF of 99.680%, and an A-20 value of 86.25%. SHAP identified normalized position as the dominant input, accounting for approximately 78.1% of the total mean absolute SHAP contribution, and reproduced the expected reduction in normalized wall deformation toward the excavation corner. SHAP and LIME further clarified the global response trends and an illustrative local prediction. The framework demonstrated strong interpolation performance within the investigated numerical domain and provides a rapid supplementary approach for reconstructing normalized wall-deflection envelopes. Independent validation using monitored excavations remains necessary before stand-alone field application.
No takes yet. Share an insight, caveat, or question.
Lei et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: