Statistical regression models for predicting the swelling pressure of compacted expansive soils

This study used 10 different soil samples to calibrate and validate two new linear regression models (M1 and M2) that use initial placement conditions and index properties to predict swelling pressure. These models were created through numerous t-tests that eliminated statistically insignificant independent variables from the final models. The absolute error of each model was calculated between the predicted swelling pressure and the actual swelling pressure values. The absolute errors of these models were then compared with the absolute errors of the models discussed in the literature that use congruent independent variables for predicting the swelling pressure. When strictly comparing the model fit measure (adjusted R-2), M2 resulted in the best fit. However, when comparing the absolute prediction error of each model, M1 appeared the most accurate of the 14 models considered. In addition, M1 is the most user friendly of the new models because it is a linear model with easily estimable independent variables for an accurate, consistent estimation of swelling pressure.