Application of Machine Learning for Factor of Safety Assessment in a Dolomite Quarry Using Geotechnical and Seismic Inputs

Authors:
Publisher website
Google Scholar
Abstract
Slope stability analysis traditionally relies on empirical classifications, analytical methods, and numerical modelling to assess the safety of rock slopes. Although these physics-based approaches provide robust results, their application in large parametric investigations and probabilistic analyses can be computationally demanding. Building upon a comprehensive geological and seismic dataset developed for the Vilonya dolomite quarry in Hungary, this study explores the capability of machine learning (ML) techniques to efficiently predict the Factor of Safety (FoS). The dataset integrates slope geometry, discontinuity characteristics, rock-mass properties derived from the Hoek–Brown and Mohr–Coulomb criteria, and seismic loading conditions representing both static and pseudo-dynamic scenarios. Four supervised learning algorithms—Random Forest, Support Vector Regression, Artificial Neural Networks, and Decision Trees—were evaluated to develop surrogate models for FoS prediction. Model performance was assessed using cross-validation to ensure robustness and generalizability. The results demonstrate that ensemble learning achieved the highest predictive accuracy, while kernel-based and neural approaches provided competitive performance. Feature importance analysis indicates that failure plane geometry and seismic loading are the primary controlling factors governing slope stability, consistent with fundamental rock mechanics principles. The findings suggest that ML-based surrogate models can significantly accelerate scenario evaluation and support probabilistic safety assessments. This data-driven framework complements conventional geotechnical methods and offers a practical tool for preliminary design, sensitivity analysis, and rapid stability screening in quarry environments subjected to seismic influence.

This publication has 10 references indexed in Scilit: