A machine learning model for predicting the unconfined compressive strength of MICP treated sands
Volume Title: ICASGE2025
Paper ID : 1053-ICASGE-FULL
Authors
New York University Abu Dhabi
Abstract
This study employs the machine learning (ML) based Multivariate Adaptive Regression Splines (MARS) model to predict the unconfined compressive strength (UCS) of Microbially Induced Carbonate Precipitation (MICP)-treated sands. The dataset includes experimental measurements, with classical predictors such as median grain size (D50), coefficient of uniformity (Cu), void ratio (e), urea concentration (Mu), calcium concentration (Mc), optical density (OD600), pH, and calcium carbonate content (Fca). The MARS model demonstrates its utility in capturing nonlinear relationships between input and output variables while maintaining interpretability through basis functions. Evaluation metrics show the MARS model achieving a root mean square error (RMSE) of 0.018 and a coefficient of determination (R²) of 0.901, highlighting its reliability in UCS prediction. The findings highlight the capability of the MARS model as an efficient and interpretable tool for UCS prediction, offering valuable guidance for optimizing MICP treatment protocols. The study underscores the effectiveness of the MARS model as a robust and interpretable tool for predicting UCS, offering a practical balance between accuracy and explainability for MICP applications.
Keywords
Subjects