ABSTRACT
The undrained shear strength is a critical design parameter in the design of soil-cement columns for ground improvement projects. This paper presents an evaluation of undrained shear strength in soil-cement columns constructed with the dry mixing method for Bangkok soft clay. The relationships between undrained shear strength and design parameters including cement content, total water-cement ratio, curing time, and natural moisture content were established using machine learning (ML) techniques in this study. Soil samples of soil-cement column (SCC) were obtained from two sites including the AIT site and Highway No. 35 site. Different varieties of cement contents were mixed using the dry method to prepare laboratory SCC samples with curing times of 7, 14, and 28 days. The SCC samples were utilized to develop models including the multivariate regression (MLR) model, random forest (RF) model, and artificial neural network (ANN) regression model for prediction of the undrained shear strength. The coefficient of determination (R2), mean absolute error (MAE), and root mean square error (RMSE) were used as performance detectors for different models. It was found that R2 values for all models ranged from 0.70 to 0.99 for the training data and from 0.5 to 0.94 for the testing data. A linear equation using the MLR model to predict undrained shear strength was proposed using cement content, natural moisture content, and curing time as the primary parameters. Various graphs to predict undrained shear strength were also proposed using the RF and ANN models.
