Sustainable Data-Driven Metaheuristic Models for Strength Evaluation of Marine Clay Treated with Recycled Tiles

As a final objective of several geotechnical experimental trials, a dependable assessment of the cohesive soils specification in conjunction with recycled materials was considered. This work utilized many attributes, including critical physical and mechanical properties of marine clay (MC) that were altered with recycled tiles (RT), to determine the unconfined compressive strength (UCS). Multiple innovative hybrid approaches were developed, integrating the least square support vector regression (LSSVR) analysis with different optimization algorithms like the golden jackal algorithm (GJA), artificial rabbit optimizer (ARA), and dwarf mongoose algorithm (DMA). Optimization techniques have been applied to the LSSVR frameworks to find the optimal hyperparameter value (LSGJA, LSARA, and LSDMA). The results indicate that the LSGJA, LSARA, and LSDMA methods have significant potential to accurately forecast the UCS. During the learning and examination stages, the R2 and indices of agreement (IA) values for the LSGJA approach were found to be, respectively, 0.992 and 0.9962 and 0.998 and 0.999. The sensitivity analysis depicted that the simulation of UCS was highly affected by all attributes higher than 0.8435 (which was related to curing time (CT)), where the highest impact belonged to maximum dry density (MDD) at 0.9594, followed by PH at 0.9524.