Rock Slope Stability Prediction: A Review of Machine Learning Techniques

Rock slope stability is a pivotal concern in geotechnical engineering, essential for mitigating risks associated with landslides and slope failures. In recent years, there has been a significant shift towards integrating machine learning (ML) techniques alongside traditional methods for enhanced analysis. Traditional Methods such as the limit equilibrium method, finite element method, and finite difference method have long served as the foundation for slope stability analysis. These methods, while well-established, face challenges in addressing the complexity of heterogeneous geological conditions and dynamic environmental factors. Furthermore, empirical systems like rock mass rating and geological strength index are often limited by subjective parameter selection, reducing their predictive reliability. Machine Learning Approaches have shown great promise in overcoming some of these limitations. ML techniques, such as convolutional neural networks, support vector machines, gradient boosting machine, Bayesian networks, random forests, and hybrid models like particle swarm optimization-artificial neural networks, can analyze large, complex datasets more efficiently. These models have been demonstrated to outperform traditional methods by incorporating real-time data, seismic activity, and environmental variability, thus enabling dynamic and real-time assessments. ML models have been shown to improve predictive accuracy for heterogeneous rock masses, facilitating better-informed decision-making in slope stability management and improving safety outcomes. This review presents a comprehensive comparison of various ML techniques, offering guidance on the selection of the most appropriate models based on specific geological conditions while highlighting their advantages. Additionally, the review highlights limitations of current ML models, reviewing real world applications and their results, which may help readers to suggest future research directions, focusing on advanced data processing methods to unlock their full potential in geotechnical engineering. This includes addressing data quality, generalization across diverse geological terrains, and computational complexity.