Determine the Value of Maximum Dry Density by Gaussian Process Regression in Individual and Hybrid Models

Abstract

This investigation introduces a novel method for utilizing the GPR technique to project the MDD within soil stability mixes. The established method is based on creating precise and thorough schemes that link a variety of natural soil characteristics, like linear shrinkage, plasticity, particle size variation, and the type as well as quantity of stabilization additives, to the MDD of stabilized soil. The schemes are created and assessed using a variety of soil types derived from earlier published destabilization outcomes from tests. Two meta-heuristic schemes, the Dingo Optimization Algorithm (DOA) and the COOT Optimization Algorithm (COA), were included in the study in this article. As a result, two hybrid schemes, consisting of GPDO and GPCO, were created. The GPCO model has a high R2 = 0.9905 value and the most propitious RMSE = 26.13 during the training stage, showing that this model has a better predictive capability and generalization ability compared to other schemes developed in this investigation. In summary, this tactic offers a viable solution for the accurate anticipation of the MDD of soil-strengthening blends employing GPR and integrating meta-heuristic schemes, which may be useful in various engineering applications.