The Implementation of a Support Vector Regression Model Utilizing Meta-Heuristic Algorithms for Predicting Undrained Shear Strength

The undrained shear strength (USS) of soil is essential in diverse structural engineering applications, including the design of earth dams, rock fills, foundations, highways, railways, and slope stability analysis. Traditional empirical and theoretical methods for estimating USS based on field tests often rely on correlation assumptions, leading to imprecise results. These conventional strategies are also limited in terms of efficiency, both in time and cost. To address these limitations, this study introduces innovative machine learning techniques, employing the Support Vector Regression (SVR) model to accurately predict USS. To enhance model performance, three meta-heuristic optimization algorithms Differential Squirrel Search Algorithm (DSSA), Golden Section Search Optimization (GSSO), and Northern Goshawk Optimization (NGO) were utilized. The frameworks were trained using four key input metrics: plastic limit (PL), liquid limit (LL), sleeve friction (SF), and overburden weight (OBW). The performance of the proposed frameworks was evaluated using five criteria: R², RMSE, MSE, SI, and SMAPE. Among the three hybrid frameworks developed for estimating USS, the SVR optimized with the Northern Goshawk Optimization (SVNG) algorithm outperformed the others, achieving the lowest RMSE value of 39.30 in the testing step and the highest R² value of 0.9980 in the testing step. These results demonstrate the superiority of the SVNG model in providing precise and efficient predictions of USS, surpassing conventional methods.