Stability Prediction of Excavations Reinforced with Different Geosynthetics – An Integrated Numerical and Machine Learning Approach

This study investigates the effectiveness of various geosynthetic reinforcements, including geotextiles, geogrids, and geocells, in enhancing the safety and performance of sandy soil excavations. A comprehensive methodological framework was implemented, combining three-dimensional finite-difference numerical simulations with five machine-learning predictive models using a 189-dataset to evaluate excavation stability in both unreinforced and geosynthetic-reinforced scenarios. The outcomes of the 3D numerical simulations indicated that geotextiles achieved optimal performance when installed at critical depths with a spacing of 70 cm . In contrast, geogrids provided superior reinforcement at spacings of 30-50 cm . Across all reinforcement types, the safety factor values showed a delayed onset of failure and improved load transfer, underscoring the crucial role of geosynthetics in mitigating the propagation of shear strain and lateral deformation. The machine learning analysis demonstrates that tree-based ensemble models, specifically Random Forest (R 2 =0.897) and Gradient Boosting (R 2 =0.872), yield the most reliable and stable predictions for this dataset. Both models provide unbiased and consistent results, with Random Forest displaying the most favorable residual distribution.