Prediction of Homogeneous Earthen Slope Safety Factors Using the Forest and Tree Based Modelling

This study assesses the potential of soft-computing based models i.e. Random Forest (RF), Random Tree (RT), M5P, Bagging M5P and Stochastic M5P for predicting safety factors (FS) of homogenous earthen slopes. For this purpose, a homogenous earthen slope was simulated with Slope/W software that uses the limit equilibrium method (LEM). Validation of the method was performed by comparing the calculations with accepted graphical results. For the model performance evaluation, five different statistical parameters including the coefficient of correlation (CC), root mean square error (RMSE), mean absolute error (MAE), scattering index (SI) and Nash–Sutcliffe model efficiency coefficient (NS) were used. Results showed the stochastic M5P based model performing better than other models with CC = 0.9950, RMSE = 0.0716, MAE = 0.0522, SI = 0.0405 and NS = s 0.9894 for the testing stages. The accuracy of the best performing model was confirmed by comparison with reported real FS and common methods. Another important conclusion was that Hybrid M5P-based models work better than traditional M5P-based models for predicting FS of soil slope. Results of a sensitivity analysis suggest that stability number (m) is the most influencing parameter for predicting the FS.