Machine learning-based approach for predicting the consolidation characteristics of soft soil

In recent times, large-scale infrastructural projects are being constructed on varieties of soil, especially in highly compressible marine clays and soft soil. The coefficient of consolidation (c(v)) is one of the most important technical parameters used to estimate the consolidation characteristics of the soil. The experimental laboratory techniques used to obtain c(v) are time-consuming and possess different practical limitations. In this study, a reliable method for predicting c(v) is presented based on machine learning (ML). The study considered 11 inherent soil variables, among which the least significant variables are discarded using univariate feature selection technique. Different ML models were developed like the random forest, artificial neural network, and support vector machine for nonlinear mapping of the c(v) using dimensionally reduced independent variables. Verification against experimental data demonstrates that the Random Forest model accurately predicts the c(v) (with MAE = 0.0231, MSE= 0.00148, and RMSE = 0.03854). Further, a comparative study of the proposed model is presented with available empirical equations and numerically simulated data. Moreover, the strengths and shortcomings of different ML algorithms are also discussed in detail.