Prediction of Soil–Water Characteristic Curves of Fine-grained Soils Aided by Artificial Intelligent Models

The advantages associated with the artificial intelligence technology can be exploited to reliably and reasonably predict the soil–water characteristic curves (SWCC) of fine-grained soils alleviating conventionally used cumbersome and time-consuming experimental procedures. In this paper, multivariate adaptive regression splines (MARS) are used as a tool along with the aid of phyisco-empirical model for predicting SWCCs of fine-grained soils. The key input variables for the proposed MARS model are derived from the grain-size distribution curve. The significance of key input variables in the model analyzed using two different sensitivity analyses investigations suggests that the SWCC behavior of fine-grained soils is strongly influenced by the clay content. Therefore, a relationship between the upper and the lower bound residual suction and clay content values has been developed and used in the MARS model. Based on all the derived information, a MARS-aided design method has been developed combining with widely used physico-empirical model and SWCC fitting equation, for rapid yet reliable technique for predicting SWCCs of fine-grained soils.