부산대학교 도서관

Soft soils pose significant challenges to the environment and construction of infrastructure on them owing to their distinct characteristics such as low bearing strength, high water content, low permeability, and high void ratio. The stiffness modulus of soft ground soils (Gs) is one of the major considerations while designing geo-structures. The determination of the stiffness modulus of soft ground materials such as soils requires expensive machinery, more skilled labor, and consumption of time which is contrary to the current trends of sustainable development. Therefore, this paper presents the artificial intelligence (AI)-based sustainable solutions for the estimation of Gs using artificial neural network (ANN), gene expression programming (GEP), and multiple linear regression (MLR) techniques. In this regard, 199 samples of soft soil from different locations were retrieved and tested to determine basic soil attributes such as sand content (S), fine content (FC), liquid limit (LL), plastic limit (PL), water content (w), and bulk density (d) which were used as potential indicators for computing soft ground stiffness modulus. Many statistical tests, including R-square (R2), root means square error (RMSE), and mean absolute error (MAE), were used to further substantiate the performance efficiency of computed prediction models. The findings show that the proposed models meet all accuracy-related acceptance requirements. However, ANN outperforms GEP and MLR. Further, to evaluate the specific impact of input factors, sensitivity and parametric tests were also executed.