Soil swelling is the most destructive geo-hazard recorded in history, considering its influence on the durability and stability of structures. The purpose of this study is to predict the one-dimensional swelling potential of soil through the Least Absolute Shrinkage and Selection Operator (LASSO) regression, Ridge regression, and Multiple Linear Regression (MLR). The database of soils extracted from literature was obtained by means of predictive laboratory experiments, including plasticity index (Ip), dry density (γd), plastic limit (PL), liquid limit (LL), and clay. The novelty of this study lies in the utilization of Lasso and Ridge regression models in soil swell potential prediction. The model’s performance metrics were mean absolute percentage error (MAPE), mean deviation (MD), root mean square error (RMSE), and coefficient of determination (R2). The results demonstrate that Lasso outperformed the other techniques with a performance accuracy of R2 values of 0.872, 0.746, 0.935, 0.931 at RMSE values of 0.106%, 0.167%, 0.046%, 0.051%. The results of this work have practical implications for engineers and advance soil mechanics when addressing swelling potential areas and machine learning applications in geotechnical engineering, to overcome the rampant laboratory visitation in search of soil data needed for slope stability, ground improvement, earthworks, and seismic analysis. Specifically, when dealing with swell potential areas, the study’s predictive assistance in making better-informed decisions regarding the construction and design phases.
Bility et al. (Tue,) studied this question.