Hybrid random forest-based models for predicting shear strength of structural surfaces based on surface morphology parameters and metaheuristic algorithms

The prediction of shear strength between soil-structure interactions is of great significance to the stability of geotechnical engineering. In this study, 480 morphological data with seven morphological parameters (deviation of the root mean square value of the profile (Pq), skewness of the height distribution in the profile (Psk), kurtosis of the height distribution of the profiles (Pku), average width of outline elements (PSm), root mean square slope of the profile (Pdq), material ratio of the profile(Pmr), number of peaks (Ppc)) were selected to generate a comprehensive database for predicting the peak interface efficiency (IEp) considering three different soil particle sizes (0.35 mm, 0.53 mm, and 0.80 mm). Three random forest (RF) models optimized using dragonfly algorithm (DA-RF), sparrow search algorithm optimized random forest (SSA-RF), and whale optimization algorithm (WOA-RF) were generated to predict IEp. and compared the predictive performance with extreme learning machine (ELM), support vector regression with radial basis function kernel (SVR-RBF) and initial RF models. The mean absolute error (MAE), the mean absolute percentage error (MAPE), the root mean square error (RMSE), and the coefficient of determination (R2) were used to evaluate the performance of all models. The results showed that the WOA-RF model has achieved the best performance by resulting in MAE of (0.0145, 0.0181, 0.0179 and 0.0210, 0.0273, 0.0216), MAPE of (1.9866, 2.6417, 2.5310 and 2.8924, 4.0294, 3.0816), and RMSE of (0 0178, 0.0237,0.0224 and 0.0252, 0.0362, 0.0276), R2 (0.9473, 0.9262, 0.9352 and 0.9404, 0.8433, 0.9313) in the training and testing phases. The results of significance analysis indicated that Pdq and Pq have more importance than other parameters for predicting IEp.

Automated summary

In this study, predictive models for shear strength in soil-structure interactions were generated using different algorithms, and important morphological parameters were selected. The results showed that the random forest model optimized using the whale optimization algorithm performed the best in terms of predictive performance.