Prediction of shallow foundation settlement: usage of hybridized support vector regression model

Because of the intricacy of cohesion soil texture, settlement simulation in cohesion materials is a critical subject to address. According to the literature study, shallow foundation prediction did not much considered, so developing optimized models can enhance the forecast accuracy in this field. It is the focus of this research to put newly developed machine learning models, such as hybridized support vector regression (SVR) with firefly algorithm (FFA) and cuckoo optimization algorithm (COA), into practice as efficient approaches to predict settlement (S-m) of shallow foundations over cohesion soil properties. The footing width, footing pressure, footing geometry, count of SPT blow, and footing embedment ratio are chosen as estimation parameters. The use of optimization techniques served the aim of determining the ideal value for the major variables of the investigated model. In the COA-SVR system, the values of R-2 and RMSE in the learning phase are 0.9649 and 4.7693, suitable than ANFIS-PSO by 0.9025 and 8.09, and in the examining phase are 0.9937 and 2.5485, considerably proper compared to ANFIS-PSO at 0.739 and 14.10, respectively. By considering another metric-like PI index, the COA-SVR network results more properly than the FFA-SVR model, with a decline of 0.0206 and 0.0717 in the learning and examining data sets, respectively. Furthermore, VAF index also depicts the same trend with outperforming the COA-SVR to FFA-SVR, especially in the examining stage, with a rise of about 3.2%. In conclusion, it is clear that the COA-SVR system could perform better than those integrated with FFA, as well as ANFI-PSO, where the proposed system can be known as the proposed system in the estimation procedure of shallow foundation S-m.