Development of a framework for the prediction of slope stability using machine learning paradigms

Accurate slope stability prediction is of utmost importance to reduce disastrous effects of slope failures and landslides. However, conventional methods of slope stability analysis are complex and challenging, and more importantly, use of these methods in a wide-area slope stability assessment requires a large number of soil property and field investigation data. These complexities and challenges often demand some simplified statistical slope stability analysis models such as by using machine learning (ML) techniques. So, in this research, we develop slope stability prediction models using multiple linear regression (MLR) and artificial neural network (ANN) and classify the slopes as safe or unsafe using random forest (RF) and support vector machine (SVM) methods. For this purpose, a dataset of 4,208 slope cases was created using limit equilibrium-based Slide software. The effectiveness of each model was then evaluated using statistical metrics and validated through roadside slope cases in Nepal, India, Canada, and the UK. In this study, Spencer’s method-based ANN model was found to have demonstrated the highest reliability. The findings of this work may contribute to simplified and better decision-making process in slope stability assessment, slope safety enhancement, and sustainability improvement in engineering projects involving soil slopes.