Assessment of Unconfined Compressive Strength of MICP-Treated Sand Using Computational Approaches

Microbially induced calcite precipitation (MICP) is a biomediated soil improvement technique that enhances sand strength through microbial calcite deposition. The unconfined compressive strength (UCS) of MICP-treated sand is a key performance indicator of this process. In this study, a deep neural network (DNN) model was developed to predict UCS and benchmarked against four machine learning models: linear regression (LR), decision tree (DT), random forest (RF), and support vector regression (SVR). A comprehensive database of 443 UCS test results was compiled from laboratory experiments and literature, incorporating geotechnical, biological, and chemical input parameters, including mean grain size, uniformity coefficient, initial void ratio, urease activity (UA), urea concentration, calcium concentration, treatment cycles (TC), and calcium carbonate content (CCC). Model performance was assessed using mean absolute error (MAE), root mean square error (RMSE), R2, and extended robustness metrics (VAF, RSR, NMBE, A20, SI, and Willmott’s d-index). The results showed that the DNN consistently outperformed all other models, achieving the highest predictive accuracy and reliability. Furthermore, SHapley Additive exPlanations (SHAP) analysis confirmed the dominant influence of CCC, TC, calcium concentration, and UA on UCS predictions, providing mechanistic interpretability of the model outputs. These findings demonstrate that DNN-based models are powerful and interpretable tools for predicting UCS of MICP-treated sand, with strong potential for supporting design and optimization in bio-mediated ground improvement.