A Machine Learning Framework for Evaluating Mechanical Performance of MICP and EICP Treated Expansive Soils

Two most popular bio-stimulated methods of soil stabilization are Microbial Induced Calcium Carbonate Precipitation (MICP) and Enzyme Induced Calcium Carbonate Precipitation (EICP). These methods are more eco-friendly compared to conventional methods of stabilization of expansive soils using cement and lime. No models are available to combine different experimental results. This paper proposes the application of machine learning to evaluate and predict the behavior of MICP and EICP treated soils by combining different experimental results. We reviewed more than twenty papers to combine the results. Machine Learning models are built using properties of MICP and EICP treated soils. It concerns most important geotechnical, physicochemical, calcite content, curing time and admixtures such as Unconfined Compressive Strength (UCS), Splitting Tensile Strength (STS). We trained the model by using 20 samples based on the Random Forest (RF) algorithm. Finally, the machine learning model was evaluated with two techniques: coefficient of determination, RMSE and MAE for regression-based and classification models respectively. The RF model achieved 75% predicting accuracy. It also had a high precision (0.75) and recall (1.00) regarding the strength improvement based on calcite content, confirming that the strengthening occurred via calcite deposition. The fact that UCS, CaCO3 content and microstructural properties (SEM/XRD) correlate very highly with each other was confirmed by correlation analysis. These results confirm the reliability of ensemble learning for stabilization trend identification and a scalable data-driven decision support system for soil engineering.