Metaheuristic machine learning based factor of safety prediction of MSE walls

Mechanically stabilized earth (MSE) walls are widely used earth-retaining structures, and their stability depends on soil-structure interaction, wall height, and the number of reinforcement layers. A 6 m high MSE wall with concrete facing panels was analyzed using PLAXIS 2D with 15-noded plane strain elements. Four input parameters-wall height, number of layers, soil cohesion, and internal friction angle were used to predict the factor of safety (FOS) through various machine learning (ML) models, including XGBoost, CART, ANN, XGBoost with LR, CART with LR, ANN with XGBoost, and a metaheuristic hybrid model integrating ANN, XGBoost, and Particle Swarm Optimization (PSO). Parametric analysis confirmed satisfactory performance of the wall configuration, with FOS increasing with cohesion and internal friction, but decreasing with wall height and number of layers. The metaheuristic hybrid model demonstrated the highest predictive accuracy (R²_train = 0.999, R²_test = 0.998). Feature importance analysis identified wall height as the most dominant factor negatively affecting FOS, followed by the number of layers. Cohesion and internal friction, on the other hand, positively contributed to stability. Statistical evaluation using the Friedman test (χ² = 18.32, p = 0.0009) and Nemenyi post-hoc analysis confirmed that the ANN + XGBoost + PSO model is statistically superior. The study highlights the effectiveness of hybrid ML models in reliably predicting FOS for MSE walls under varying conditions.