Application of improved ANFIS approaches to estimate bearing capacity of piles

An accurate estimation of deep foundation bearing capacity in different types of soils with the aid of the field experiments results is taken into account as the most important problems in geotechnical engineering. In recent decade, applying a broad range of artificial intelligence (AI) models has become widespread to solve various types of complicated problems in geotechnics. In this way, this study presents an application of two improved adaptive neuro-fuzzy inference system (ANFIS) techniques to estimate ultimate piles bearing capacity on the basis of cone penetration test (CPT) results basically used in analysis of pile foundations. The first model was combination of ANFIS and group method of data handling (GMDH) and the second one was related to the integration of fuzzy polynomial (FP) and GMDH model. Furthermore, in the ANFIS–GMDH, constant coefficients of ANFIS model were optimized using gravitational search algorithm (GSA). To improve the proposed AI models for carrying out training and testing stages, a reliable database in form of four input variables included information about different properties of soils and driven piles obtained from CPTs results. Performance of the proposed approaches indicated that FP–GMDH had better performance (RMSE = 0.0647 and SI = 0.378) in comparison with ANFIS–GMDH–GSA (RMSE = 0.084 and SI = 0.412). The use of multiple linear regression and multiple nonlinear regression equations showed lower level of precision in prediction of axial-bearing capacity of driven piles compared to the ANFIS–GMDH–GSA and FP–GMDH techniques.