A novel Harris hawks' optimization and k-fold cross-validation predicting slope stability

Stability of the soil slopes is one of the most challenging issues in civil engineering projects. Due to the complexity and non-linearity of this threat, utilizing simple predictive models does not satisfy the required accuracy in analysing the stability of the slopes. Hence, the main objective of this study is to introduce a novel metaheuristic optimization namely Harris hawks' optimization (HHO) for enhancing the accuracy of the conventional multilayer perceptron technique in predicting the factor of safety in the presence of rigid foundations. In this way, four slope stability conditioning factors, namely slope angle, the position of the rigid foundation, the strength of the soil, and applied surcharge are considered. Remarkably, the main contribution of this algorithm to the problem of slope stability lies in adjusting the computational weights of these conditioning factors. The results showed that using the HHO increases the prediction accuracy of the ANN for analysing slopes with unseen conditions. In this regard, it led to reducing the root mean square error and mean absolute error criteria by 20.47% and 26.97%, respectively. Moreover, the correlation between the actual values of the safety factor and the outputs of the HHO-ANN (R-2 = 0.9253) was more significant than the ANN (R-2 = 0.8220). Finally, an HHO-based predictive formula is also presented to be used for similar applications.

Automated summary

The study introduced a novel Harris hawks' optimization (HHO) method to enhance the accuracy of the conventional multilayer perceptron technique in predicting the safety factor of slopes with rigid foundations, by adjusting computational weights of conditioning factors. The use of HHO significantly increased the prediction accuracy of ANN for analyzing slopes with unseen conditions, reducing errors and showing higher correlation with actual safety factor values.