Reliability analysis of reinforced soil slope stability using GA-ANFIS, RFC, and GMDH soft computing techniques

Abstract

Soil is a heterogeneous medium, the characteristics that determine soil slope stability are highly variable, making the analysis a difficult task. The present research approach is switching from deterministic to probabilistic in order to account for the variability in soil properties. This research presents the use of three soft-computing techniques to evaluate reinforced soil slope reliability based on slope stability: Genetic Algorithm based Adaptive Network based Fuzzy Inference System (GA-ANFIS), Random Forests Classifier (RFC), and Group Method of Data Handling (GMDH). Shear strength parameters c (cohesion), ϕ (angle of shearing resistance) and ϒ (unit weight) are used as input variables, while Factor of Safety of Reinforced Soil Slope (F) is used as an output variable to determine the stability of a soil slope of a certain height. The Models were also evaluated using various assessment parameters and GA-ANFIS outperformed having some testing outputs as NS= 0.997, RMSE= 0.017, VAF= 99.731 %, Bias Factor= 1.002, PI= 1.998, R2 = 0.997, GPI= 6.6E-08, U95 = 0.627, tstat= 0.247 and β = 1.543. The GA-ANFIS model outperformed the GMDH and RFC models, according to the findings of the analyses using Taylor diagram and ROC curve. As a result, the GA-ANFIS model can be utilized as a reliable soft computing technique to analyze reinforced soil slope stability.