Predictive Modelling for Slope Stability Analysis by Machine Learning: A Comprehensive Review

Abstract

Slope stability assessment is essential for ensuring the safety and sustainability of civil engineering projects such as dams, highways, and excavations. The conventional analysis procedures, such as limit equilibrium and numerical simulation, are usually insufficient to handle the complex as well as the nonlinear characteristics of soil and rock masses under varying environmental circumstances. The paper will present a critical review of recent developments in the field of predicting slope stability with the help of machine learning (ML), deep learning (DL), and hybrid optimization. It is a summary of the empirical evidence of different state-of-the-art studies on such algorithms as Support Vector Machines (SVM), Random Forest (RF), Gradient Boosting Machines (GBM), Artificial Neural Networks (ANN), Multivariate Adaptive Regression Splines (MARS) and deep architectures like Long Short-Term Memory (LSTM) and Convolutional Neural Networks (CNN). Classical regression measures (R², MAE, RMSE), classification measures (accuracy, ROC-AUC), and model interpretability measures (SHAP values) are presented in the performance assessment, providing a helpful comparison of the predictive performance and feature importance. The fact that nowadays the metaheuristic optimizers (e.g., Particle Swarm Optimization, Firefly Algorithm) can be combined with the ML algorithms is highlighted in terms of the enhancement of model accuracy and robustness. In the work, the researchers show that geotechnical parameters, including those related to cohesion, slope height, and internal friction angle, are of essential importance on the slope failure. There are also restrictions such as data sparsity and the threats of overfitting, which are discussed. Future research directions of the paper are encompassed with the incorporation of sensor networks, data-driven hybrid models, and explainable artificial intelligence methods with the view to instilling adaptive and dependable slope stability assessments.