Predicting the effective stress parameter of unsaturated soils using adaptive neuro-fuzzy inference system

The effective stress parameter (χ) is applied to obtain the shear strength of unsaturated soils. In this study, two adaptive neuro-fuzzy inference system (ANFIS) models, including SC-FIS model (created by subtractive clustering) and FCM-FIS model (created by Fuzzy c-means (FCM) clustering), are presented for prediction of χ and the results are compared. The soil water characteristic curve fitting parameter (λ), the confining pressure, the suction and the volumetric water content in dimensionless forms are used as input parameters for these two models. Using a trial and error process, a series of analyses were performed to determine the optimum methods. The ANFIS models are constructed, trained and validated to predict the value of χ. The quality of the ANFIS prediction ability was quantified in terms of the determination coefficient (R2), Root Mean Square Error (RMSE) and Mean Absolute Error (MAE). These two ANFIS models are effectively able to predict the value of χ with reasonable values of R2, RMSE and MAE. Sensitivity analysis was used to acquire the effect of input parameters on χ prediction, and the results revealed that the confining pressure and the volumetric water content parameters had the most influence on the prediction of χ.