Abstract
The nature of soil varies horizontally as well as vertically, owing to the process of the formation of soil. Thus, ensuring the safe design of geotechnical structures has been a major challenge. In shallow foundations, conducting field tests is expensive and time-consuming and often conducted on significantly scaled-down models. Empirical models, too, have been found to be the least reliable in the literature. The study proposes AI-based techniques to predict the bearing capacity of a shallow foundation, simulated using the datasets obtained in experiments conducted in different laboratories in the literature. The results of the ELM-EO and ELM-PSO hybrid models are compared with that of the ELM and MARS models. The performance of the models is analyzed and compared with each other using various performance parameters. The models are graded to each other using rank analysis and the visual interpretations are provided using error matrices and REC curves. ELM-EO is concluded to be the best performing model (R2 and RMSE equal to 0.995 and 0.01, respectively, in the testing phase), closely followed by ELM-PSO, MARS, and ELM. The performance of MARS is better than ELM (R2 equals 0.97 and 0.5, respectively, in the testing phase); however, hybridization greatly enhances the performance of the ELM and the hybrid models perform better than MARS. The paper concludes that AI-based models are robust and hybridization of regression models with optimization techniques should be encouraged in further research. Sensitivity analysis suggests that all the input parameters have a significant influence on the output, with friction angle being the highest.