Seismic Slope Stability Prediction Using an Optimized Deep and Cross Network: An AI-Based Approach

This paper aims to propose slope stability based on the critical unit weight of AUS clay while considering a linearly increasing undrained shear strength subjected to horizontal seismic acceleration forces occurring from earthquake events, utilizing two-dimensional finite element limit analysis (FELA) and an advanced hybrid framework combining the Deep and Cross Network (DCN) with the slime mould algorithm (SMA). The SMA model is utilized to optimize the hyperparameters of the DCN to enhance its predictive performance using a total dataset of 3,360 samples obtained from FELA, with 70% allocated for training and 30% for testing. The parametric studies employed to analyse the dimensionless stability number ( N ) of this problem include the slope angle ( β ), horizontal seismic acceleration coefficient ( k h ), depth‒height ratio of the soil slope ( d / H ), anisotropic strength ratio ( r e ), and strength gradient factor ( ρH / S uc0 ). The results of this study were proposed in the form of a relationship chart design between d / H , ρH / S uc0 , r e , β , k h , and the N value. Additionally, the failure mechanisms of shear dissipation and comparisons between existing results are presented in this study. A surrogate model developed from DCN-SMA is carried out by employing FELA results as a dataset. Furthermore, the performance assessment using statistical metrics—R 2 , RMSE, and MAE—demonstrated that the DCN-SMA model delivers outstanding accuracy on both the training and testing datasets, with values of 0.966976, 0.485919, and 0.362970 for the training set, and 0.967038, 0.481985, and 0.363763 for the testing set, respectively. In addition, the SHAP and CFI results indicate that ρH/S uc0 is the most influential parameter in both the FELA and DCN-SMA models, followed by k h , β , r e , and d/H , respectively. • Proposing a seismic slope stability framework for AUS clay considering linearly increasing undrained shear strength using finite element limit analysis (FELA) • Developing a hybrid framework combining the deep and cross network (DCN) with the slime mould algorithm (SMA) to predict the slope stability factor under horizontal seismic loading • Investigating the effects of slope angle, seismic acceleration coefficient, depth–height ratio, anisotropic strength ratio, and strength gradient factor on slope stability • Demonstrating excellent predictive performance of the DCN–SMA surrogate model with R 2 values exceeding 0.96 on both training and testing datasets