<p>The current research evaluates the performance of the presented models in forecasting the uplift resistance factor (F_c) of circular anchors in clays that are both anisotropic and heterogeneous. The numerical data from prior scientific research, which was made public, were used to train the machine learning models. To achieve this goal, the Least Squares Support Vector Regression (LSSVR) algorithm was designed. To locate the appropriate set of parameters, the LSSVR is linked to the processes of the Coati Optimization (CO) and the Mountain Gazelle Optimizer (MGO), as hyperparameters are of utmost importance in this simulation. Three dimensionless parameters are used to identify the uplift resistance. These parameters are the normalized outcomes parameter, the ratio of embedment, strength, and anisotropy. The uplift resistance is then stated regarding the uplift resistance factor F_c. The findings suggest that in anisotropic, heterogeneous clays, there is a strong likelihood that LSSVR, LSSVR (CO), and LSSVR (MGO) will provide accurate estimates of the F_c. The projected performance measures clearly show that LSSVR (MGO) is a more accurate and reliable approach than LSSVR and LSSVR (CO). The difficulty of precisely calculating the uplift resistance of circular anchors in anisotropic and heterogeneous clays is addressed in this work. The suggested models produced extremely accurate predictions (R² up to 0.997) with the lowest error indices using LSSVR optimized with CO and MGO algorithms. The LSSVR–MGO framework is the most dependable and efficient way to anticipate uplift resistance, according to the results, and it performs better than other methods.</p>