The stability of natural slopes plays a vital role in ensuring the integrity and safety of civil engineering infrastructures, such as roads, bridges, and buildings. Traditional methods for assessing slope stability often face significant limitations, including challenges in accurately determining the mechanical parameters of the soil and rock and high computational demands that can prolong the analysis process. In recent years, machine learning-based techniques have emerged as promising alternatives, offering more efficient and reliable solutions for slope stability assessment.
In this study, five distinct machine learning-based methods were employed to predict the factor of safety against slope failures. The methods analyzed include Gaussian Process Regression (GPR), Multi-Layer
Perceptron (MLP), Simple Linear Regression (SLR), Support Vector Regression (SVR), and Multiple Linear Regression (MLR). To develop these models, a comprehensive dataset was created, consisting of 504 training instances and 126 testing instances. This dataset was generated through finite limit equilibrium analysis modelling, which provides a foundational understanding of the stability conditions of slopes.
To evaluate the efficiency and predictive capability of the models, various statistical indices were employed. After rigorous testing, the results indicated that the Multi-Layer Perceptron (MLP) outperformed all other machine learning models, achieving an impressive total score of 50, which reflects its superior accuracy in predicting slope stability. Meanwhile, the Support Vector Regression (SVR) and Multiple Linear Regression (MLR) models also demonstrated commendable accuracy, both securing total scores of 35. Conversely, the Gaussian Process Regression (GPR) and Simple Linear Regression (SLR) models exhibited lower performance, with total scores of 20 and 10, respectively, indicating their acceptable yet less reliable accuracy in this context.
The findings from this study underscore the effectiveness of machine learning-based methods as a viable solution for assessing the factor of safety against slope failures. The Multi-Layer Perceptron model’s prominence in predictive accuracy suggests that it should be prioritized in practical applications. Furthermore, the encouraging results from the SVR and MLR models indicate that they can serve as reliable options as well. Ultimately, this research advocates for the adoption of machine learning techniques as promising alternatives to traditional methods for slope stability assessment, potentially
enhancing the safety and efficiency of civil engineering practices.
Reference
Tien Bui, D., Moayedi, H., Gör, M., Jaafari, A., & Foong, L. K. (2019). Predicting slope stability failure through machine learning paradigms. ISPRS International Journal of Geo-Information, 8(9), 395.