Explainable Machine Learning for Soilcrete UCS Predictions

Soil mixing is an in-situ soil treatment method that consists of mixing cementitious binders with in-situ soil to create soilcrete. The success of this treatment often relies on the value of the Unconfined Compressive Strength (UCS) of the soilcrete after a given curing time. Estimating the UCS can be challenging because numerous factors influence the results, including soil type, soil moisture content, binder content, soil chemistry, soil heterogeneity, and mixing means and methods. In current North American design-build practice, pre-construction design UCS values are typically estimated qualitatively based on a contractor’s experience. Existing correlations may not be suitable for current applications because of recent advances in deep mixing equipment and methodologies. There is a need for more rational and quantitative estimates of UCS for use in pre-construction design analyses. This paper will explore the intersection of Machine Learning (ML) with geotechnical engineering and soilcrete applications. A database of soilcrete UCS and site/soil/means/methods metadata is compiled from recent Keller projects and leveraged to explore UCS prediction with advanced ML regression techniques. From this ML exploration, a blueprint of how to scaffold, feature engineer, and prepare soilcrete data for various ML techniques will be created. However, many ML models sacrifice explainability for higher accuracies. To achieve insights from ML models, Explainable ML will be applied to the ML models to explain variable importances. Explainable ML can reveal complex patterns and interactions in the model variables, along with the relative importance of each variable’s contributions to the final UCS value. The insights received from the Explainable ML model can then be further pursued in traditional geotechnical research approaches to expand soil mixing knowledge.