Effective stress parameter in unsaturated soils; an evolutionary-based prediction model

Deformations and failures in unsaturated soils are influenced directly by the effective stress calculated using the stress equation affected by the effective stress parameter. A data mining-based approach, the Evolutionary Polynomial Regression (EPR), is implemented in this research to develop a prediction model for the effective stress parameter in unsaturated soils. The proposed modelling approach takes an evolutionary computing technique to for finding polynomial models that are structured and explicit. A combination of the well-established genetic algorithm method and the least square approach are implemented to search for the most suitable polynomial structures and their corresponding parameters for all terms in the developed polynomial structure. A set of unsaturated soil experimental results (triaxial tests) from literature were used in this study to develop the prediction model. Once the model completed it was evaluated based on its performance for making predictions using input parameters that were previously kept unseen to validate generalization capabilities (making predictions of the output for new input data). The predictions made by the model, were compared to actual measured data from the lab tests as well as an Artificial Neural Network based model. A sensitivity analysis was also done to assess the level and form of contributions that input parameters had to the developed model. The results showed that the developed model could successfully and to a high level of accuracy capture and redevelop the intrinsic connections between the input parameters involved in the model to help produce accurate the effective stress parameter predictions that can not only compete with the artificial neural network model in terms of accuracy of the model predictions and generalisation capabilities; but also outperform the artificial neural network model with regards to the structure, simplicity and transparency.