Linear regression to predict the unconfined compressive strength of biopolymer-based soil treatment (BPST)

Abstract

Biological soil treatment methods have recently been actively promoted for sustainable and ecofriendly geotechnical engineering. Biopolymer-based soil treatment (BPST) is recognized as a low-carbon footprint ground improvement approach with appropriate pore clogging and strengthening properties. BPST is typically applied as a combination of soil, biopolymer, and water; however, depending on the hydrogel phase and soil type, varied results can be produced. In this study, Decision Tree, a machine learning approach, was used to predict the unconfined compressive strength (UCS) of BPST. The model performed successfully, and the determinant accuracy was more than R2=0.99. And through permutation feature importance, it was confirmed that biopolymer content and water content act as determinants for the prediction of UCS of BPST.