Unified model for prediction of the strength of soils stabilized with xanthan gum using white-box machine learning

Biopolymer-based soil stabilization has attracted increasing attention as a sustainable approach in geotechnical engineering. However, accurate prediction of the compressive strength of biopolymer-stabilized soils remains challenging due to the complex interaction among soil properties, additive characteristics, and curing conditions. In this study, a comprehensive database comprising 248 unconfined compressive strength (UCS) tests on soils stabilized with xanthan gum was compiled and analyzed, and a unified and interpretable predictive framework applicable to a wide range of soil types and curing conditions was developed. To derive an explicit numerical relationship, a white-box machine learning algorithm based on the Group Method of Data Handling (GMDH) was employed to predict UCS. The input parameters included plasticity index (PI), xanthan gum content (XGC), curing time (CTi), curing temperature (CTe), initial moisture content (WMC), and the unconfined compressive strength of untreated soil (q₀). The developed model demonstrated strong predictive capability for the compiled dataset, achieving an overall coefficient of determination of R² = 0.883. When evaluated on an independent testing subset, the model yielded an R² value of approximately 0.75 with an RMSE of 0.62 MPa, indicating satisfactory generalization performance. Sensitivity analysis using both SHAP-based global interpretation and classical approaches indicated that intrinsic soil properties and curing-related variables govern UCS, with the unconfined compressive strength of untreated soil and plasticity index identified as the most influential parameters. These findings provide a quantitative prediction tool and physically consistent insight into the behavior of xanthan gum-stabilized soils, offering practical guidance for engineering applications.