Predicting the Settlement of Shallow Foundation Using Metaheuristic SVR Approaches

Abstract

Because cohesive soil structure is so complicated, settlement modelling is, in some ways, essential. The goal of this study is to identify settlement ($S_{m}$) of shallow foundations using newly developed machine learning methods, such as hybridized support vector regression ($SVR$) with sine–cosine algorithm ($SCA$) and Bat-inspired algorithm ($BAT$). Footing width, pressure, geometry, number of standard penetration tests, and footing embedment ratio are among the estimated variables. The application of optimization methods served the objective of identifying the optimal value for the primary variables of the researched model. The SCA–SVR is thought to be the best framework with the highest classification when compared to BAT–SVR and ANFIS–PSO. During the process of learning, the values of $R^{2}$ and $MAE$ are 0.9629 and 1.5354, which is preferable to ANFIS–PSO by 0.9025 and 4.92, and 0.9823 and 1.3787 in the evaluating portion, which is superior to ANFIS–PSO at 0.739 and 9.88. By looking at $PI$ indicator, the SCA–SVR network does better than the BAT–SVR in both the education and evaluating datasets. The SCA–SVR model dropped 0.0262 and 0.0435 points in the education and evaluating data sets, respectively. Likewise, the $A_{10 - index}$ index exhibits a similar tendency. In conclusion, after analyzing the accuracy and taking into account the definitions, it is completely obvious that the $SVR$ combined with $SCA$ can work better than $BAT$, as well as by literature, that could be referred to as the proposed system in the forecasting model of $S_{m}.$