Estimating Deformation of Geogrid-Reinforced Soil Structures Using Hybrid LSSVR Analysis

The study of displacement plays a key role in the planning of geosynthetic reinforced soil structures. Nevertheless, the literature stresses the promise of artificial intelligence technologies in tackling geotechnical engineering difficulties. The major purpose of this work was to evaluate the potential utilize of machine learning-based approaches in forecasting the deformation of geogrid-reinforced soil structure (Dis). This study introduces and verifies novel techniques that integrate the reptile search algorithm (RSA) and equilibrium optimizer (EO) with least squared Support Vector Regression (LSSVR). Afterward, a total of 166 finite element analyses conducted in the literature were used in order to create the dataset. The aim of the application of optimization algorithms was to find the optimal values of the penalty factor (c) and the width (g) of the kernel function for LSSVR. The results show that both the \(LSSVR_E\) $LSSV{R}_E$ and \(LSSVR_R\) $LSSV{R}_R$ algorithms have a good chance of correctly forecasting the \(Dis\) $\mathit{Dis}$ . Considering the \(TIC\) $\mathit{TIC}$ index, a remarkable reduction was concluded, a reduction from 0.0393 \(\left( {LSSVR_E } \right)\) $\left(LSSV{R}_E\right)$ to 0.0215 \(\left( {LSSVR_R } \right)\) $\left(LSSV{R}_R\right)$ in train phase, and from 0.0222 \(\left( {LSSVR_E } \right)\) $\left(LSSV{R}_E\right)$ to 0.0088 \(\left( {LSSVR_R } \right)\) $\left(LSSV{R}_R\right)$ in the test phase. A comprehensive index named \(OBJ\) $\mathit{OBJ}$ concluded 1.8003 for \(LSSVR_E\) $LSSV{R}_E$ , and an almost a half reduction at 0.9257 for \(LSSVR_R\) $LSSV{R}_R$ .