Prediction of soil compaction parameters through the development and experimental validation of Gaussian process regression models

The laboratory determination of maximum dry density (ρ_dmax) and optimum moisture content (w_opt) of soils requires considerable time and energy. Efforts have been made in the past to present models to predict the soil compaction parameters (ρ_dmax and w_opt), but the existing models are either applicable to specific soil types, plasticity range, compaction energy, or they have low prediction accuracy. This study aims to develop novel prediction models of soil compaction parameters using Gaussian Process Regression (GPR) incorporating different soil types, plasticity ranges, and compaction energies. The database used to develop prediction models consists of the index properties and compaction parameters of soils. GPR models were developed based on different kernel functions. To assess the accuracy of the predictive models, various error metrics were used, including coefficient of determination (R²), mean absolute error, and root mean square error. The validity of the models was verified by conducting laboratory experiments, where satisfactory results were obtained for the prediction of compaction parameters. A comparison of the performance of the proposed models in this study was also made with the existing models in the literature which showed that the models presented in this research performed better than those in the literature both during the model development and also in experimental validation. Finally, a sensitivity analysis was performed which indicates that the prediction models are greatly affected by plastic limit and fine contents.