Linear and nonlinear approaches and statistical evaluations to predict the shear strength parameters and collapse potential of gypseous soils

Abstract

Gypsum soils suffer from engineering problems of collapse, which occurs when an increase in moisture causes significant soil deformation in the constructions. There is no sufficient information about the most important physical soil characteristics that affects the shear strength parameters and collapse potential of the gypseous soil, which covers about 30% of the area of Iraq. The present research illustrates the efficiency of the linear and nonlinear regression to model the shear strength parameters; cohesion (c), and angle of internal friction (∅), together with compressibility characteristics; and compression index (Cc) and collapse potential (CP) of gypseous soils based on a wide range of gypsum content (0 to 77.48%) and physical soil properties. The cohesion, angle of internal friction, compression index, and collapse potential ranged from 0 to 260 kPa, 3 to 54.5°, 0.055 to 0.91, and 0.19 to 20.38%. A wide range of 220 collected data points from different academic published researches were used to develop the constitutive models. According to adjusted (R2), mean absolute error (MAE), and the root mean square error (RMSE), the c, ∅, and CP of gypseous soil are well predicted in terms of gypsum content, initial water content, void ratio, liquid limit, plasticity index, total unit weight, and dry unit weight. Based on the statistical assessments of higher R2, lower MAE, and RMSE, the models were confirmed to predict the c, ∅, Cc, and CP of the gypseous soils. The sensitivity analysis of the models showed that the liquid limit and total unit weight are the most influential parameters in determining c and ∅, respectively, with the training dataset, while the most significant physical soil properties in estimating the CP were the ratio of specific gravity to the void ratio.