Abstract:
The maximum dry density (MDD) and optimum moisture content (OMC) are the two important compaction parameters that are obtained using proctor tests in the laboratory, but they require energy and time. Therefore, extensive work has been done in the literature to predict these parameters rather than actually performing the proctor tests in the laboratory but either the developed models are applicable to specific soil type, specific compaction energy or the performance of the method, in terms of accuracy, is compromised when dealing with large dataset. In this study, three machine learning methods; Gene expression programming (GEP), Artificial Neural Network (ANN) and Gaussian Process Regression (GPR), were used to develop prediction models for soil compaction parameters; maximum dry density (MDD) and optimum moisture content (OMC) with higher accuracy. The database used to develop the prediction models was obtained from the literature. The dataset consists of both fine grained and coarse-grained soils; soils ranging from low plasticity to high plasticity; and compacted using different compaction energies. The performance of the developed models was evaluated based on coefficient of determination (R2), mean absolute error (MAE), and root mean square error (RMSE) and a comparison was made with the Multi Expression Programing (MEP) model from the literature for the same database. It was found that all the new prediction models from this study performed better than the MEP model. In terms of R2, ANN performed much better as compared to GEP and GPR. All the three developed models, in different forms, can be used to predict the compaction parameters for new datasets.
