Machine learning methods to map stabilizer effectiveness based on common soil properties

Most chemical stabilization guidelines for subgrade/base use unconfined compressive strength (UCS) of treated soils as the primary acceptance criteria selecting optimum stabilizer in laboratory testing. Establishing optimal additive content to augment UCS involves a resource-intensive trial-and-error procedure. Also, samples collected from discrete locations trials may not be representative overall site. This study aims minimize number and help strategize sampling by developing spatial maps at different treatment levels lime cement. These were developed using machine-learning techniques, database compiled various reported studies on cement United States. Supervised learning methods under regression classification categories used quantify classify values after treatments, respectively. Commonly available soil properties like Atterberg limits, gradation, organic contents along with type amount predictors response. Median R2 best model was 0.75 0.82 cement, while Correct Prediction Rate (CPR) 92% 80% Results showed that satisfactory predictions could made regarding effectiveness simple information commonly available. Best performing models selected generating two counties Montana. Soil these tested verify predictions. The results indicate Pearson’s correlation coefficient 0.78 CPR 92%. authors hope this future will increase data-driven-decision-making geotechnical engineering practices.