Abstract
This study develops a least absolute shrinkage selection operator (lasso) for realistic three-dimensional (3D) geotechnical problems. We propose a lasso-based method specialized for subsurface modeling considering the difference between the vertical and horizontal correlations of soil properties. The straightforward computational implementation of this geotechnical lasso (GLasso) is not feasible in 3D problems because of significant runtime and memory exceedance. To address this problem, we propose an efficient implementation of the alternating direction method of multipliers (ADMM). GLasso and the efficient ADMM (eADMM) were applied to a real case study of subsurface modeling to demonstrate their performance in geotechnical practice.